Research lines of the Doctoral Programme in Network and Information Technologies

Explainable artificial intelligence (AI) focuses on making the decision-making processes of AI systems, such as deep learning models, transparent and understandable to humans. As AI systems become increasingly powerful and widely used across multiple domains, their "black box" nature poses significant challenges for trust, accountability, and effective human oversight. The goal of explainable AI is to provide techniques that help understand how the AI systems represent the information, and how the information is used by the AI systems to produce outputs.

The field of explainable AI emerged around 2014, right after the beginning of the deep learning revolution [1,2]. Since then, several techniques have been developed to improve model interpretability. For example, feature attribution methods, such as LIME (Local Interpretable Model-Agnostic Explanations) [3] and SHAP (SHapley Additive exPlanations) [4], estimate the contribution of each input feature to a model's prediction. In turn, visualization methods, such as CAM [5], are post hoc explainability methods for computer vision models that produce heatmaps to visualize the regions of the input data (images) that contributed the most to the output. Other techniques focus on revealing information of the internal representation learned by the model [6]. More recently, autonomous agents that actively probe models have emerged as a methodology of automating interpretability, enabling model analysis at scale. For instance, the Multimodal Automated Interpretability Agent (MAIA) [7] framework enables agents to perform iterative experimentation with specific multimodal tools, while the Self-Reflective Interpretability Agent (SAIA) [8] framework adds a self-reflection mechanism that allows the agent to revise faulty hypotheses based on experimental evidence, leading to more accurate and robust conclusions.

The use of one technique or another really depends on the purpose and context of the interpretability analysis, since different objectives, data modalities, and stakeholders require different forms of explanations [9]. For example, in model debugging and validation, feature attribution may help developers identify spurious correlations within the model and the data. Revealing spurious correlations is necessary to ensure alignment between the model behaviour and human expectations, which is essential in sensitive application domains such as healthcare. In contrast, automated interpretability agents offer scalable solutions for continuous model monitoring and evaluation in complex and dynamic environments.

This research line on explainable AI will focus on two complementary directions:
1. Application of existing explainability techniques to diverse domains, such as medical image analysis or interpretability of multi-modal models to support clinical decisions.
2. Development of novel explainability methods that go beyond current approaches in interpretability, with the ultimate goal of ensuring robustness and alignment with human expectations.

Bibliography

[1] B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba. "Object Detectors Emerge in Deep Scene CNNs." International Conference on Learning Representations (ICLR). 2015.

[2] Q. Zhang, Z. Song-Chun, "Visual interpretability for deep learning: a survey", Frontiers of Information Technology & Electronic Engineering, 2018.

[3] M. T. Ribeiro, S. Singh, C. Guestrin, in Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, USA 2016, pp. 1135–1144

[4] S. M. Lundberg, S.-I. Lee, Advances in Neural Information Processing Systems, The MIT Press Number 12, Vol. 30, 2017.

[5] B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, A. Torralba. Learning Deep Features for Discriminative Localization, CVPR, 2016.

[6] D. Bau, J-Y Zhu, H. Strobelt, A. Lapedriza, B. Zhou, A. Torralba, "Understanding the role of individual units in a deep neural network", Proceedings of the National Academy of Sciences (PNAS), 2020.

[7] Tamar Rott Shaham, Sarah Schwettmann, Franklin Wang, Achyuta Rajaram, Evan Hernandez, Jacob Andreas, and Antonio Torralba. A multimodal automated interpretability agent. In Forty-first International Conference on Machine Learning, 2024.

[8] C. Li, J.Lopez-Camuñas, J. Thomas, J. Andreas, A. Lapedriza, A. Torralba, T. Rott, "Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent", Neural Information and Processing Systems (NeurIPS), 2025.

[9] Ribera, Mireia, and Agata Lapedriza. "Can we do better explanations? A proposal of user-centered explainable AI." IUI Workshops. 2019.

Facial expressions are a key source of information for developing new technologies. Humans use facial cues to convey emotions, and psychologists have studied this since Charles Darwin's early work [1]. One of the most influential models is the Facial Action Coding System (FACS) [2], which defines action units (facial muscle movements) as the basis for six basic emotions: happiness, surprise, fear, anger, disgust and sadness. Understanding this near-universal language is a major focus in computer vision, with applications in human-computer interaction. However, human emotions go far beyond this basic set. Our research applies deep learning to enable computers to interpret emotions from facial and prosodic cues.
We emphasize applications in child psychology, educational tech and e-health, where multimodal data can support personalized learning and mental health diagnostics.

We explore innovative uses of advanced technology in mixed research designs with children in natural settings (e.g., schools). Our work includes developing ethical, efficient and non-intrusive methods to collect and process audiovisual data, as well as applying statistical and spatio-temporal models to analyse machine-generated data, including coding and compression techniques.

This PhD thesis will be conducted in collaboration with the Child Tech Lab and the AI for Human Well-being Lab at the UOC.

[1] Darwin, Charles (1872), The expression of the emotions in man and animals, London: John Murray.
[2] P. Ekman and W. Friesen. Facial Action Coding System: A Technique for the Measurement of Facial Movement. Consulting Psychologists Press, Palo Alto, 1978.

In recent years, end-to-end learning algorithms have revolutionized many areas of research, such as computer vision [1], natural language processing [2], gaming [3], robotics and protein folding [4]. Deep learning techniques have achieved the highest levels of success in many of these tasks, given their astonishing capability to model both the features/filters and the classification rule.

Despite these advancements, there remains considerable potential for enhancing deep learning methodologies to better address real-world challenges. This research initiative will explore novel approaches in the following areas:

- Federated learning: a machine learning technique that enables training deep learning models on decentralized data sources. This approach addresses privacy concerns and logistical challenges associated with data sharing, making it particularly valuable in healthcare and other privacy-sensitive domains.

- Uncertainty estimation and conformal predictions: The output of a Neural Network typically consists of a score indicating the probability or regression of a user-defined label. However, real-world applications necessitate explicit quantification of the uncertainty associated with these predictions. For instance, in a cancer diagnosis system, it is crucial to provide not only the type and probability of cancer but also the confidence level of the prediction [5]. To address this, we propose the development of novel deep learning methodologies that explicitly model predictive uncertainty. Furthermore, we aim to leverage this uncertainty to facilitate active and semi-supervised learning on extensive unlabelled datasets.

These algorithms will be applied to real computer vision problems in the fields of automated medical diagnosis in e-Health applications, or automated identification, tracking and population analysis in several citizen science projects. We collaborate with several hospitals in the Barcelona area and two research centres where these methods are currently applied: VHIR (Vall d'Hebron Institut de Recerca), CSIC-ICM (Marine Sciences Institute https://www.icm.csic.es/en).

[1] A. Krizhevsky, et al. Imagenet classification with deep convolutional neural networks. In NIPS, 2012.
[2] Sutskever, I., et al. Sequence to sequence learning with neural networks. In Advances in neural information processing systems. 2014
[3] Silver, D., et al. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484-489.
[4] Gal, Y., Ghahramani, Z.: Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. In: ICML. (2016).
[5] Adhane, G., Dehshibi, M. M., and Masip, D. (2021). A Deep Convolutional Neural Network for Classification of Aedes Albopictus Mosquitoes. IEEE Access, 9, 72681-72690.

We are already beginning to interact daily with intelligent machines – social robots, virtual agents and other autonomous systems – that are becoming part of our homes, workplaces and public spaces. These technologies are increasingly capable of sustaining long-term relationships with people [1]. Ongoing research and deployment efforts are demonstrating their potential in diverse domains, such as socially assistive systems that support healthcare [2], robots that aid and accompany older adults [3,4] and intelligent agents that enrich teaching and learning experiences [5].

An important aspect for these intelligent machines to communicate fluently with people is the capacity perceiving expressions of emotions, preferences, needs or intents, and reacting to those expressions in a socially and emotionally intelligent manner. This research line focuses on designing and implementing technologies that allow intelligent machines to have these types of simulated emotional intelligence, which are essential abilities to maintain social interactions with humans.

More generally, this area of research includes other subareas related with the development of emotional intelligence in machines, like emotion perception in the wild [6,7], analysis of emotions and social cues in dyadic interactions, text sentiment analysis [8], visual sentiment analysis [9] and emotion perception from audio [10]. In terms of applications, we focus on human assistance, human companionship, entertainment and emotional well-being.

References:

[1] C. Kidd and C. Breazeal (2008). "Robots at Home: Understanding Long-Term Human-Robot Interaction". Proceedings of the 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2008). Nice, France.

[2] Breazeal, C. (2011, August). Social robots for health applications. In 2011 Annual international conference of the IEEE engineering in medicine and biology society (pp. 5368-5371). IEEE.

[3] Broekens, J., Heerink, M., & Rosendal, H. (2009). Assistive social robots in elderly care: a review. Gerontechnology, 8(2), 94-103.

[4] Camuñas, J.L., Bustos, C., Zhu, Y., Ros, R. and Lapedriza, A., 2025. Experimenting with Affective Computing Models in Video Interviews with Spanish-Speaking Older Adults. In Proceedings of the Winter Conference on Applications of Computer Vision (pp. 1-10).

[5] Belpaeme, T., Kennedy, J., Ramachandran, A., Scassellati, B., & Tanaka, F. (2018). Social robots for education: A review. Science Robotics, 3(21).

[6] R. Kosti, J.M. Álvarez, A. Recasens and A. Lapedriza, "Context based emotion recognition using emotic dataset", IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2019.

[7] Cabacas-Maso, J., Ortega-Beltrán, E., Benito-Altamirano, I., & Ventura, C. (2025). Enhancing Facial Expression Recognition with LSTM through Dual-Direction Attention Mixed Feature Networks and CLIP. In Proceedings of the Computer Vision and Pattern Recognition Conference (pp. 5665-5671).

[8] Felbo, B., Mislove, A., Søgaard, A., Rahwan, I., & Lehmann, S. (2017). Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm. arXiv preprint arXiv:1708.00524.

[9] C. Bustos, C. Civit, B. Du, A. Sole-Ribalta, A. Lapedriza, "Leveraging Vision-Language models for Visual Sentiment Analysis: a study on CLIP", 11th International Conference on Affective Computing & Intelligent interactions (ACII), 2023.

[10] Ortega-Beltrán, E., Cabacas-Maso, J., Benito-Altamirano, I., & Ventura, C. (2024, September). Better Spanish Emotion Recognition In-the-wild: Bringing Attention to Deep Spectrum Voice Analysis. In European Conference on Computer Vision (pp. 335-348). Cham: Springer Nature Switzerland.

Oculomics is the study of how the health of the eye can reflect the general health of the body, using the eye as a "window" to detect and understand systemic diseases. This field integrates advanced imaging technologies and artificial intelligence to identify biomarkers in the eye that may indicate the presence of diseases like diabetes, cardiovascular diseases [1,2], anaemia [3], kidney disease [4], Alzheimer's [5] and other conditions. The goal is to enable earlier (even before clinical symptoms appear), non-invasive and cost-effective diagnoses of diseases affecting the rest of the body, significantly reducing the economic burden for healthcare systems.

This research project will focus on analysing traditional (fundus, OCT or retinal angiography) and advanced (such as AOSLO) retinal imaging, developing deep learning models that are able to early diagnose for preventive treatment, thereby facilitating new, more personalized medicine approaches and prevention plans.
The three main challenges when dealing with these applications are:

1. Small sample size problems, usually the N is reduced, and the generalization capabilities of the network are affected. We will explore transfer learning and the use of generative model methods for this purpose.

2. The resulting models should be explainable and easy to interpret. We will provide both a classification score and an explanation of this score, to make the early diagnosis more reliable and trustable.

3. Federated learning (FL). Due to the sensitivity and regulations (like HIPAA/GDPR) surrounding patient health data, sharing is difficult. The healthcare sector is increasingly adopting federated learning (FL) to overcome this by sharing model parameters (knowledge) instead of data, maintaining privacy with a decentralized machine learning approach.

The resulting methods will be transferred to hospitals from the Barcelona metropolitan area, and the research efforts will result in a strong social return.

[1] Gerrits et al. (2020). Age and sex affect deep learning prediction of cardiometabolic risk factors from retinal images. Scientific Reports, 10(1), 1-9.
[2] Barriada, R. G., Simó-Servat, O., Planas, A., Hernández, C., Simó, R., & Masip, D. (2022). Deep Learning of Retinal Imaging: A Useful Tool for Coronary Artery Calcium Score Prediction in Diabetic Patients. Applied Sciences, 12(3), 1401.
[3] Tham, Y. C., et al. (2020). Detection of anaemia from retinal images. Nature Biomedical Engineering, 4(1), 2-3.
[4] Sabanayagam, C., et al. (2020). A deep learning algorithm to detect chronic kidney disease from retinal photographs in community-based populations. The Lancet Digital Health, 2(6), e295-e302.
[5] McGrory et al. (2017). The application of retinal fundus camera imaging in dementia: a systematic review. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, 6, 91-1

The rise of large vision-language models (VLMs), such as CLIP [1], ALIGN [2], Flamingo [3], GPT-4V [4] and Gemini [5], has moved computer vision from traditional classification tasks towards performing complex high-level vision-language reasoning tasks that were previously out of reach. These new VLMs can be used for video understanding, visual question answering, detailed image/video captioning, and even open-ended dialogue grounded in visual content.

This research line on large deep learning multimodal models aims to bridge vision systems with cognitive capabilities by leveraging and extending current VLMs and also by developing novel multimodal models that incorporate additional modalities beyond vision and text (e.g. audio or physiological signals). We focus on developing multimodal deep learning architectures to address high-level vision and language tasks, such as:

- Emotion perception: Given an image or a video, building models that can perceive emotions represented in the visual content [6].
- Visual question answering (VQA): Given an image or video and a question, building models that generate an appropriate answer [7].
- Image/video captioning: Automatically generating detailed and context-aware textual descriptions of visual content [8].
- Image/video segmentation: Identifying and segmenting specific regions or objects in an image or video based on a description [9].
- Inclusive VLMs: Analysing the extent to which current VLMs are inclusive across demographic, cultural and linguistic dimensions. This line of research investigates biases in data and model outputs, evaluates accessibility and representation in multimodal systems, and develops strategies for building more equitable and socially responsible VLMs (e.g. [10]).

Bibliography

[1] Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., ... & Sutskever, I. (2021, July). Learning transferable visual models from natural language supervision. In International conference on machine learning (pp. 8748-8763). PmLR.

[2] Masry, A., Rodriguez, J. A., Zhang, T., Wang, S., Wang, C., Feizi, A., ... & Rajeswar, S. (2025). Alignvlm: Bridging vision and language latent spaces for multimodal understanding. arXiv preprint arXiv:2502.01341.

[3] Alayrac, J. B., Donahue, J., Luc, P., Miech, A., Barr, I., Hasson, Y., ... & Simonyan, K. (2022). Flamingo: a visual language model for few-shot learning. Advances in neural information processing systems, 35, 23716-23736.

[4] OpenAI. Gpt-4v(ision) technical work and authors. 2023.

[5] Gemini Team, Google. Gemini: A Family of Highly Capable Multimodal Models. 2023. Available online: https://gemini.google.com

[6] C. Bustos, C. Civit, B. Du, A. Sole-Ribalta, A. Lapedriza, "Leveraging Vision-Language models for Visual Sentiment Analysis: a study on CLIP", 11th International Conference on Affective Computing & Intelligent interactions (ACII), 2023.

[7] Khan, Z., BG, V. K., Schulter, S., Yu, X., Fu, Y., & Chandraker, M. (2023). Q: How to specialize large vision-language models to data-scarce vqa tasks? A: Self-train on unlabeled images! In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 15005-15015).

[8] Hu, X., Gan, Z., Wang, J., Yang, Z., Liu, Z., Lu, Y., & Wang, L. (2022). Scaling up vision-language pre-training for image captioning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 17980-17989).

[9] Rasheed, H., Maaz, M., Shaji, S., Shaker, A., Khan, S., Cholakkal, H., ... & Khan, F. S. (2024). Glamm: Pixel grounding large multimodal model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 13009-13018).

[10] S. Dudy, I.S. Ahmad, R. Kitajima, A. Lapedriza, "Analyzing Cultural representations of Emotions in LLMs through Mixed Emotion Survey", 12th International Conference on Affective Computing & Intelligent interactions (ACII), 2024. (Best Paper Award).

While deep learning has enhanced interactive video retrieval [1], automated multicamera editing remains underexplored. This research line builds on our recent work [2], which addresses automated classical concert editing by decomposing the problem into two key sub-tasks: "when to cut" (temporal segmentation) and "how to cut" (spatial shot selection).

For "when to cut," we developed a lightweight, multimodal convolutional-transformer (audio, time features, visual embeddings) that outperforms statistical baselines. For "how to cut," we improved shot selection by replacing older backbones with a CLIP-based encoder for better semantic alignment, trained on a pseudo-labeled dataset created using a hybrid pipeline including LLM-based confirmation (Gemini).

Future research will focus on:
- Extending the temporal model to a regression task to predict precise cut timestamps.
- Incorporating higher-level affective cues, like musical emotion or visual emotion, to guide shot selection.
- Developing prompt-based or user-in-the-loop editing systems.
- Applying these models to new domains (e.g., sports, stage plays, public speaking).
- Implementing long-term temporal modeling for globally coherent editing.

These developments underscore the potential for automated systems to handle complex multicamera scenarios, applying learned stylistic cues and viewer-centric strategies to produce engaging video content.

Bibliography

Lokoč, J., Andreadis, S., Bailer, W., et al. Interactive video retrieval in the age of effective joint embedding deep models: lessons from the 11th VBS. Multimedia Systems, 29(6), 2023, pp. 3481-3504.

Gonzálbez-Biosca, D., Cabacas-Maso, J., Ventura, C., & Benito-Altamirano, I. (2025, October). When and How to Cut Classical Concerts? A Multimodal Automated Video Editing Approach. In Proceedings of the 3rd International Workshop on Multimedia Content Generation and Evaluation: New Methods and Practice (pp. 98-106).

Wu, H. Y., Santarra, T., Leece, M., Vargas, R., and Jhala, A. Joint attention for automated video editing. Proceedings of the 2020 ACM International Conference on Interactive Media Experiences, 2020, pp. 55-64

The increasing energy consumption and significant privacy concerns inherent in traditional deep-learning vision systems necessitate a fundamental change in camera design. Instead of relying on exhaustive remote computing for every detection task, this research line explores the design of minimalist cameras that "compile" the vision algorithm directly into the camera's hardware [1,2].

This is accomplished through the principles of neuromorphic optical computing[1]. Specifically, the approach involves designing a trainable optical mask that is placed over a reduced array of pixels (photodiodes) [2]. This mask is then trained using deep learning techniques as the first layer of a neural network , contrasting with older compressed imaging methods that focused on image reconstruction. This methodology is explored in the current literature on Minimalist Vision with Freeform Pixels and builds on the concept of The Minimalist Camera [3].

The integration of the algorithm directly into the hardware has profound benefits, including a reduced carbon footprint, minimized dependence on large data centers, and a significant improvement in user privacy. The technology is a powerful tool not only for classical vision problems but also could be for high-impact applications in health and planetary well-being, such as biomedical or agri-food microscopy. Previous work has demonstrated the potential of cost-effective microscopy solutions for tasks like yeast cell detection and count in the wine value chain [4,5].

A PhD proposal in this field could tackle the design, fabrication and application of minimalist cameras.

Bibliography:

[1] Kalinin, K. P., Gladrow, J., ... Ballani, H. (2025). Analog optical computer for AI inference and combinatorial optimization. Nature, 645(8080).

[2] Klotz, J., & Nayar, S. K. (2025). Minimalist Vision with Freeform Pixels. Computer Vision – ECCV 2024.

[3] Pooj, P., Grossberg, M., Belhumeur, P. N., & Nayar, S. K. (2018). The Minimalist Camera. BMVC, 141.

[4] Benito-Altamirano, I., Moreno, S., Vaz-Romero, D. M., ... Diéguez, Á. (2025). MicroVi: A Cost-Effective Microscopy Solution for Yeast Cell Detection and Count in Wine Value Chain. Biosensors, 15(1), 40.

[5] Takhar, D., Laska, J. N., ... Baraniuk, R. G. (2006). A new compressive imaging camera architecture using optical domain compression. Computational Imaging IV, 6065.

The integration of colorimetry with computer vision, employing both traditional and deep learning methodologies, has provided significant advancements in transforming qualitative sensors into precise colorimetric measurement tools. This transformation is crucial for industrial and environmental applications, where accurate color detection can offer valuable information. In 2021, we presented a practical approach for printed sensor labels capable of detecting various gases in the ambient air through colorimetric changes, highlighting the potential of such tools for broader applications in different industries [1].

Machine-readable patterns, such as QR codes, can be used in colorimetric tasks as a means to encode and transmit color reference information or calibration data, thus enhancing the efficiency of image capture and analysis processes. We demonstrated how modified QR codes could be applied in challenging conditions to convey colorimetric data, ensuring robust performance in diverse environments [2, 3]. This work provided a foundational understanding of how traditional computer vision can be integrated with deep learning to handle complex imaging tasks effectively.

Key areas of ongoing research in this field include:

- Color Correction Algorithms: Leveraging deep learning to enhance color correction processes, leading to more consistent color reproduction across various sensors and devices [1].

- Enhanced Image Capture Techniques: Utilizing deep learning models, such as those applied in QR code reading, to improve the accuracy and reliability of color data extraction [4].

- Multimodal Image Fusion: Employing techniques that combine data from various imaging modalities (e.g., RGB-D, RGB-T, IR-RGB) to expand the capabilities of classification and segmentation networks, as demonstrated by advances in multispectral detection [5].

These advancements signal a future where qualitative sensors can become powerful, reliable tools in colorimetric applications, empowering industries with accessible and precise measurement solutions.

Bibliography

1. Engel, L., Benito-Altamirano, I., Tarantik, K. R., Pannek, C., Dold, M., Prades, J. D., & Wöllenstein, J. Printed sensor labels for colorimetric detection of ammonia, formaldehyde and hydrogen sulfide from the ambient air. Sensors and Actuators B: Chemical, 2021.

2. Benito-Altamirano, I., Martínez-Carpena, D., Casals, O., Fàbrega, C., Waag, A., & Prades, J. D. Back-compatible Color QR Codes for colorimetric applications. Pattern Recognition, 133, 2023, 108981.

3. Benito-Altamirano, I., Martínez-Carpena, D., Lizarzaburu-Aguilar, H., Fàbrega, C., & Prades, J. D. Reading QR Codes on challenging surfaces using thin-plate splines. Pattern Recognition Letters, 2024.

4. Shi, B., Wang, X., Lyu, P., Yao, C., & Bai, X. Robust scene text recognition with automatic rectification. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 4168-4176.

5. Qingyun, F., Dapeng, H., & Zhaokui, W. Cross-modality fusion transformer for multispectral object detection. arXiv preprint arXiv:2111.00273, 2021.

Raman spectroscopy is a powerful non-destructive analytical technique relying on the inelastic scattering of monochromatic light, usually from a laser, by molecules in the sample. The interaction of the incident photons with the molecular vibrations causes a shift in the energy of the scattered light, known as the Raman shift. This shift provides a unique spectral fingerprint that is characteristic of the molecular structure, chemical composition and physical state of the material, making it an invaluable tool across various scientific disciplines, including chemistry, materials science and biomedical research.

This thesis proposal aims to develop novel deep learning algorithms for the automated, objective interpretation of spectral signatures [1]. Although it may have multiple applications, a primary application, currently supported by ongoing collaborations with clinicians in the Barcelona area, focuses on precise tumour classification and prognosis. This work aims to provide clinicians with a safe, convenient and easily interpretable method, thereby replacing subjective manual interpretation with objective analysis.

The group currently collaborates with a multidisciplinary network of experts: from ocular oncology clinicians (HUB-IDIBELL, HSJD), to biophotonics researchers (ICFO, HSJD), who will contribute with technical knowledge and clinical data for supporting this research.

[1] Terán, M., Ruiz, J. J., Loza-Álvarez, P., Masip, D., & Merino, D. (2025). Open Raman spectral library for biomolecule identification. Chemometrics and Intelligent Laboratory Systems, 105476.

MoCoTo (Mobile Corneal Topographer) allows for low-cost corneal topographies, using an add-on for a mobile phone or tablet, improving the adaptation of contact lenses to corneal shape.
The add-on illuminates the cornea of the eye with a specific pattern, and this is used to determine the shape of the cornea. This information can be used to adapt the shape of contact lenses and improve the experience of the user.

The analysis of the back of the eye (fundus) is crucial to identify not only eye disease but also serious neurological conditions that can lead to blindness, brain injury and even death [1]. Determining these conditions is a common clinical challenge in ophthalmic, neurosurgical and neurological clinics, and their accurate assessment sometimes requires invasive testing or surgical procedures that carry moderate to severe risks of pain and injury. Given these risks, non-invasive alternatives are imperative [2,3,4,5].

At present, eye screening image and video data (fundal photography, OCT, etc.) require interpretation by an expert assessor, limiting their usefulness in non-specialist environments. AI-driven automated tests would help guide the need for urgent referral in community settings or the appropriate follow-up investigations and interventions in hospital settings.

We aim to develop automated, deep learning-enabled image and video detection systems to create novel tests for eye and brain disease diagnosis and forecasting. The study is conducted by an interdisciplinary and international team of ophthalmology, neurology and artificial intelligence experts.

Keywords: Deep learning, explainability, medical imaging, retina, vascular disease, brain disease.

[1] Ptito, M., Bleau, M., & Bouskila, J. (2021). The retina: a window into the brain. Cells, 10(12), 3269.
[2] Panahi, A., Rezaee, A., et al. (2023). Autonomous assessment of spontaneous retinal venous pulsations in fundus videos using a deep learning framework. Scientific Reports, 13(1), 14445.
[3] Cheung, C. Y., Wong, W. L. E., et al. P. (2022). Deep-learning retinal vessel calibre measurements and risk of cognitive decline and dementia. Brain Communications, 4(4), fcac212.
[4] Cheung, C. Y., Ran, A. R., et al. (2022). A deep learning model for detection of Alzheimer's disease based on retinal photographs: a retrospective, multicentre case-control study. The Lancet Digital Health, 4(11), e806-e815.
[5] Zee, B., Wong, Y., et al. (2021). Machine-learning method for localization of cerebral white matter hyperintensities in healthy adults based on retinal images. Brain Communications, 3(3), fcab124.

The variety of imaging devices and techniques used in ophthalmology departments hinders the implementation of deep learning systems (DLSs) in eye clinics, as the systems are not agnostic to the imaging device. The poor generalizability of deep learning models without access to large amounts of unbiased data is well documented and is often the case in clinical practice, given the limited number of available cases of pathological eyes compared to controls.

Screening algorithms also fail to achieve clinically acceptable levels of performance when used on images of the same modality but acquired in less favourable conditions [1]. However, DLSs developed under such conditions can reach clinically acceptable performance if trained on the appropriate source data [2]. This poor generalizability and adaptability, together with the poor interpretability of the current models, limits the translation of successful DLSs from research to the eye clinic, where data are more variable and multimodal. This is a major issue that affects not just eye care but is common throughout frontline clinical practice.

A potential solution is to use generative models to synthesize or modify images from various modalities and conditions, which can in turn be used to train DLSs to detect diseases that are invariant to imaging conditions and devices [3,4]. Moreover, by exploring the latent space of generative models, model causality can be identified and linked to comprehensive scene attributes and model interpretability.

This project aims to develop synthetic models that generate and modify retinal images to improve disease assessment, and also to explore the causality and interpretability of generative models to extract comprehensive knowledge about morphology and lesions.

Keywords: deep learning, GAN, diffusion, medical imaging, retina, vascular disease.

[1] Nunez do Rio, J. M., Nderitu P., et al. (2022). Evaluating a deep learning diabetic retinopathy grading system developed on mydriatic retinal images when applied to non-mydriatic community screening. Journal of Clinical Medicine, 11(3), 614.
[2] Nunez do Rio, J. M., Nderitu P., et al. (2023). Using deep learning to detect diabetic retinopathy on handheld non-mydriatic retinal images acquired by field workers in community settings. Scientific Reports, 13(1), 1392.
[3] Waisberg, E., Ong, J., et al. (2025). Generative artificial intelligence in ophthalmology. Survey of ophthalmology, 70(1), 1-11.
[4] Remtulla, R., Samet, A., et al. (2025). A Future Picture: A Review of Current Generative Adversarial Neural Networks in Vitreoretinal Pathologies and Their Future Potentials. Biomedicines, 13(2), 284.

A broad range of 2 and 3-dimensional imaging devices are available in hospital care and increasingly in the community [1]. Increased high-resolution and high-contrast imaging techniques with different acquisition technologies provide practitioners with a varied spectrum of image data.

A comprehensive patient assessment often involves not just an evaluation of the patient characteristics. In most cases, multiple imaging modalities and views are used to arrive at the correct diagnosis [2,3,4]. Given the myriad imaging devices, for the full potential of deep learning (DL) impact in health care to be realized, exploitation of multimodal DL approaches and feature integration is required. DL systems (DLSs) stand to benefit from a better view of pertinent characteristics that may be exclusive to specific imaging procedures, but which in aggregate could significantly increase the accuracy of multi-modal DLSs in detecting disease and predicting its progression.

In this project we aim to develop deep learning-driven systems for the automated detection of medical conditions using different image and video data views and modalities.

Keywords: Deep learning, multiview, multimodal, MRI, CT, X-ray, mammography, echocardiography, PET.

[1] Abhisheka, B., Biswas, et al. (2024). Recent trend in medical imaging modalities and their applications in disease diagnosis: a review. Multimedia Tools and Applications, 83(14), 43035-43070.
[2] Liu, D., Gao, et al. (2022). Transfusion: multi-view divergent fusion for medical image segmentation with transformers. MICCAI (pp. 485-495). Cham: Springer Nature Switzerland.
[3] Zhao, Z., Hu, J., Zeng, Z., et al. (2022). Mmgl: Multi-scale multi-view global-local contrastive learning for semi-supervised cardiac image segmentation. In 2022 IEEE international conference on image processing (ICIP) (pp. 401-405). IEEE.
[4] Van Tulder, G., Tong, Y., & Marchiori, E. (2021). Multi-view analysis of unregistered medical images using cross-view transformers. MICCAI (pp. 104-113). Cham: Springer International Publishing.

Alzheimer's disease (AD) is the main neurodegenerative dementia that primarily affects memory. Consequently, AD typically leads to progressive deficits in memory, thinking, and daily functioning. The disease manifests in several stages, ranging from mild cognitive impairment due to Alzheimer's pathology to severe dementia. Studying AD progression remains challenging, particularly in the early stages. Early and accurate diagnoses and studying their progression are essential for optimal patient management and care planning. Magnetic resonance imaging (MRI) plays a vital role in supporting diagnosis and monitoring disease progression by detecting characteristic patterns of brain atrophy. Sleep parameters have been increasingly recognized as both potential biomarkers and contributing factors in AD progression.

Our project aims to address this critical gap in medical research. Using advanced data from MRI, we aim to study AD progression and how it interacts MRI data with sleep parameters. Furthermore, we would use sleep parameters for early diagnosis of AD and how it can help to study its progression. We will use explainable machine learning techniques and normative models to track disease progression at the individual level and to explore the different progression rates.

This innovative research will be conducted in collaboration with the Alzheimer's Disease and Cognitive Disorders Group at Barcelona's IDIBAPS-Hospital Clínic, a world-renowned clinical institution. Together, we aim to make significant strides in understanding AD progression, ultimately improving patient outcomes.

Frontotemporal dementia (FTD) is a neurodegenerative dementia that primarily affects the frontal and temporal lobes of the brain. These regions are typically associated with personality, behaviour, language and decision-making. FTD typically leads to changes in these functions. Thus, FTD presents various subtypes, such as behavioural FTD (bvFTD) and primary progressive aphasia (PPA). Diagnosing FTD, however, remains a challenge due to the overlap of symptoms with other neurodegenerative and psychiatric conditions. Achieving an early and accurate diagnosis is crucial to improve patient care. Magnetic resonance imaging (MRI) plays an important role in diagnosing and tracking the disease's evolution.

Our project aims to address this critical gap in medical research. Using advanced data from structural MRI and diffusion tensor imaging (DTI), we aim to classify FTD patients with the help of statistical methods and machine learning algorithms. The classification information, using explainable machine learning techniques, will allow us to identify potential biomarkers to help diagnose the various FTD subtypes. Moreover, by leveraging MRI and DTI data, we plan to create normative models to track disease progression on an individual basis and to explore the different progressions according to FTD subtype.

This innovative research will be conducted in collaboration with the Alzheimer's Disease and Cognitive Disorders Group at Barcelona's IDIBAPS-Hospital Clínic, a world-renowned clinical institution. Together, we aim to make significant strides in understanding and diagnosing FTD, ultimately contributing to better outcomes for patients

Agentic artificial intelligence (agentic AI) represents a new paradigm in artificial intelligence research. It focuses on developing systems that can autonomously perceive, reason, plan and act to perform a complex task [1]. In contrast with traditional AI models, which are trained to perform just one task, replicating a certain input-output behaviour, agentic AI has a certain degree of autonomy, pro-activity and adaptability, and is characterized by its capacity for goal-oriented behaviour and iterative decision-making. Agentic AI is reshaping the landscape of intelligent systems, offering the potential for adaptive solutions across multiple domains, including healthcare and education, among others.

This research line on agentic AI will explore architectures and applications of agentic AI, and has the following potential directions:

- Novel architectures: developing new mechanisms to integrate perception, reasoning, memory and planning into unified agent frameworks capable of long-term and adaptive behaviour.
- Self-reflective agents: developing systems that can evaluate their own reasoning processes, detect errors or biases, and refine their strategies [2,3].
- Human-AI collaboration [4]: designing agents that can interact and cooperate effectively with humans, adapting their communication style and level of autonomy.
- Domain-specific applications: applying agentic AI to critical domains such as medical decision support [5], or AI model interpretability and debugging [3,6].

Bibliography

[1] Acharya, D.B., Kuppan, K. and Divya, B., 2025. Agentic ai: Autonomous intelligence for complex goals–a comprehensive survey. IEEE Access.

[2] N. Shinn, F. Cassano, A. Gopinath, K. Narasimhan, and S. Yao. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36:8634–8652, 2023.

[3] C. Li, J. Lopez-Camuñas, J. Thomas, J. Andreas, A. Lapedriza, A. Torralba, T. Rott, "Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent", Neural Information and Processing Systems (NeurIPS), 2025.

[4] Nicoletti, Bernardo, and Andrea Appolloni. "A digital twin framework for enhancing human–agentic AI–machine collaboration." Journal of Intelligent Manufacturing (2025): 1-17.

[5] Karunanayake, N., 2025. Next-generation agentic AI for transforming healthcare. Informatics and Health, 2(2), pp.73-83.

[6] J.Lopez-Camuñas, C. Li, T. Rott, A. Torralba, A. Lapedriza, "OpenMAIA: a Multimodal Automated Interpretability Agent based on open-source models", Workshop on Mechanistic Interpretability at Neural Information and Processing Systems (NeurIPS Workshops), 2025

There are currently various bottlenecks in the growth in parallel and distributed programming paradigms and environments, which are affecting the ability to provide efficient applications for performing concurrent computations.

We need to know the platforms, their performance, the underlying hardware and networking technologies, and we must be able to produce optimized software that statically or dynamically may take advantage of the computational resources available.

In this line of research we study different approaches to producing better scientific applications, and to making tools (via automatic performance analysis), which can understand the application model and the underlying programming paradigm. We try to tune the performance of these to a dynamically changing computational environment, in which the resources (and their characteristics) can be homogeneous or heterogeneous depending on the hardware platform. In particular we focus our research on shared memory and message-passing paradigms, and in many-core/multi-core environments including multi-core CPUs, GPUs (graphic cards computing) and cluster/grid/cloud/super computing platforms.

In the current growth of HPC/Cloud/IA big data centers, there is a need for more energy efficiency and more sustainable sources of energy. In some cases, the initial focus comes in the form of an overall cost reduction, other times, the main objective is maintaining the computational scalability without excessive cost penalties. In the end, there is a need for more generic models for energy efficiency, and also for searching the best energy sources at each moment in the nowadays big computations.

In this research line, we will focus on different approaches to enhance computation to make it more energy-conscious, and more effective management in cloud/edge hybrid paradigms, mixing remote and local computations. And how to make those environments more sustainable. Also in the HPC/BigData/IA/IoT emerging industrial sectors, they have found different issues, in the form of scalability limitations (penalized by cost), that could lead to a halt in their activity if they do not find more flexible and sustainable ways to control the energy consumed.

We will analyze and design new predictive models to model computational workloads, their energy needs, and the best possible schedule to use different available energetical sources. Also making dynamical interaction models between energy markets and the current huge-scale computational needs.

We are looking for PhD candidates interested in helping to design computational systems more energy-aware and making better future models to optimize computation taking into account energy sources and their sustainability.

We work on a multi-criteria allocation mechanism for large-scale distributed systems using computers voluntarily provided by participants in the system.

This PhD proposal is aimed at improving our multi-criteria allocation mechanism. Possible research lines:
* Extending the selection criteria with new properties
* Exploring how to take into account energy efficiency and the use of clean energies in the allocation mechanism
* Improving node clustering
* Enhancing the complementarity of nodes
* Reducing overheads
* Exploring how to improve the quality of the data collected from each node (e.g. using some reputation mechanism)

Our main papers related to this allocation mechanism are as follows:
1. Multi criteria biased randomized method for resource allocation in distributed systems: Application in a volunteer computing system. 2018. Future Generation Computer Systems, Vol. 82, pp. 29-40.
https://recerca.uoc.edu/documentos/655f4daedfaaaa2236075164

It defines the basis of our multi-criteria allocation mechanism (MCBR).

2. CLARA: A novel clustering-based resource-allocation mechanism for exploiting low-availability complementarities of voluntarily contributed nodes. 2022. Future Generation Computer Systems, Vol. 128, pp. 248-264.
https://recerca.uoc.edu/documentos/655f4e20dfaaaa22360763d4

It improves the allocation mechanism by allowing groups of complementary nodes to behave as a node. To find the complementarities between the available computers, we must first predict their availability and then sort them into groups using a clustering technique.

3. NARA: Network-Aware Resource Allocation mechanism for minimizing quality-of-service impact while dealing with energy consumption in volunteer networks. 2025. Future Generation Computer Systems, Vol. 164.
https://recerca.uoc.edu/documentos/674cde9fe6f4e33b0a1a635a

It extends the multi-criteria mechanism with network information.

Internet Computing & Systems Optimization (ICSO) is an official IN3 programme supported by the DPCS research group. One of the main research topics in ICSO is the development of new hybrid algorithms and methods which combine applied optimization (eg heuristics and metaheuristics), discrete-event simulation and data analysis to support decision-making processes in realistic environments. In particular, we are interested in the real-life application of these algorithms in the contexts of logistics, transportation and production systems. Thus, the doctoral theses will be related to any of the following topics: rich (real-life) vehicle routing problems, real-life scheduling problems, green logistics, intelligent transport systems, horizontal collaboration in logistics, etc. These topics represent important challenges for the industrial sector in any developed country, which explains their relevance in the context of current international research.

Descriptive business analytics (DBA) refers to the pre-processing and processing of historical data gathered by companies in order to describe the real business context, generate information, and make rational decisions from the acquired knowledge. After describing the real business context, one can also benefit from predictive business analytics (PdBA), which relies on the use of time series analysis, regression models, and even machine learning methods in order to forecast the future or predict how certain business factors will evolve under scenarios of uncertainty. Finally, prescriptive business analytics (PsBA) aims at supporting complex decision-making processes in business through the use of optimization and simulation algorithms (including metaheuristics and simheuristics). These algorithms allow managers to “learn from the future” (by performing what-if analyses), significantly increasing the efficiency of business processes and systems, reducing operational costs and, at the end of the day, raising companies’ profits. PsBA have many applications in a variety of fields, including smart cities, transportation and logistics, finance, healthcare, tourism, etc. The aim of this line of research is to explore, from an interdisciplinary and IT-based perspective, some of the almost unlimited applications of analytics to any of the previously described service industries.

Fusion energy aims to address the rapidly growing global energy demand while reducing carbon emissions. To design and operate future fusion reactors—such as ITER and DEMO—accurate predictive models are essential. Simulation and modelling are therefore central pillars of fusion research, enabling the study of plasma behaviour, reactor performance, or material challenges in the reactor. However, fusion codes may be computationally very demanding. They typically require the use of large-scale HPC infrastructures and supercomputers. Our research group focuses on the optimisation and acceleration of fusion simulation workflows. The doctoral theses in this research line will investigate how Artificial Intelligence and techniques such as machine-learning-based surrogate modelling can be integrated into existing fusion codes to reduce execution time by orders of magnitude, while preserving physical accuracy.

The primary goal of this thesis proposal is to analyze and optimize the organizational issues concerning the transition of organizations from disjointed compliance efforts (GDPR compliance handled separately from AI Act compliance) to a single, integrated strategy that ensures ethical and legal standards are met consistently from data collection to AI deployment and monitoring. Research topics include but are not limited to: Operationalizing Ethics and Organizational Structure, Organization and Information Systems Perspective and Impacts,

Operationalizing Ethics and Organizational Structure, addressing questions such as: 1) What are the most effective organizational models for integrating accountability? This includes defining the roles, responsibilities, and reporting lines for merged functions, such as a Data & AI Governance Committee that includes technical, legal, and ethical expertise. 2) Developing  metrics that can quantify the success of a unified governance program (e.g., how much does an increase in data quality investment reduce the incidence of fairness-related harm reports?). 3) How can training and incentive structures be designed to embed a culture where AI developers naturally consider data privacy and ethical principles at the earliest stages of the data collection pipeline, rather than treating compliance as a post-development checklist?

Organization and Information Systems Perspective and Impacts, addressing questions such as: 1) Which are the minimal critical factors for AI Governance, Data Governance and a unified approach? 2) How organizations structure their resources, make decisions about AI and Data Governance investments, and align AI and Data Governance with overall business strategy? 3) Explore technology adoption and diffusion of Data and AI Governance. 4) How AI and Data Governance impact organizational structure, culture, job roles, communication, and overall performance.

We focus on the wider research area of the study of the processes involved in the full set of outsourcing and provisioning processes (analysis, selection, development, deployment, adoption and assessment) of ICT systems and services, by both private companies and public entities.

Thus, our scope covers the initial procurement decisions (i.e. deciding strategies, selecting providers, choosing existing solutions, etc.) and the process of monitoring the deployment and operation of the ICT services (i.e. relationship with providers, controlling strategies, stakeholder roles, project management issues, etc.), among others.

Our interest focuses on the analysis and development of models and methods for managing ICT services, informed by the broad principles of IT governance in the domain of strategic management and planning. That also embraces the design and implementation of technological and organizational solutions from a strategic perspective, as well as the impact on these of data science and artificial intelligence.

Data hiding encompasses techniques for embedding information within various carriers, enabling solutions for security, privacy and integrity across a broad spectrum of digital environments. While traditionally applied to digital media (images, audio and video) to address challenges in content security (such as preventing unauthorized copying) and digital forensics, the discipline has expanded significantly.

Lately, data hiding is crucial for ubiquitous computing and the internet of things (IoT), where it is applied to non-traditional data. Key emerging areas include:

1. Data provenance and traceability: Embedding metadata within sensor streams or IoT network traffic to track the origin and processing history of data, which is essential for auditability and trust.
2. Integrity and tamper detection: Using watermarking and similar techniques to guarantee the integrity of sensor data and system logs, allowing for the detection of unauthorized modifications in critical applications.
3. Combating disinformation and deepfakes: Digital watermarking is used to verify media authenticity. This involves provenance watermarking (robust) to trace media origin and fragile watermarking for detecting any subtle digital manipulation or evidence of deepfakes.
4. Intellectual property protection: The principles of watermarking extend to protecting the intellectual property of artificial intelligence (AI) models. Machine learning model watermarking is used to embed signatures within model parameters, allowing creators to claim ownership and detect unauthorized distribution or use of their proprietary models.
5. Covert communications: Using steganography to send concealed messages in seemingly innocent data streams, and steganalysis to detect and analyse such hidden communications.

This research line focuses on designing, analysing, and developing novel data hiding mechanisms, emphasizing requirements like security, capacity, robustness, transparency and computational efficiency for deployment in resource-constrained IoT devices and networked distribution systems. These topics are fundamentally related to computer forensic techniques that provide mechanisms for verifying data authenticity and establishing legal evidence.

The Internet of things (IoT) refers to the internetworking of devices (including smartphones), vehicles, embedded systems, sensors, actuators, and other hardware and software components, which enable these objects to collect and exchange data. These data can be used later on (or in real time) for a wide variety of applications. For example, samples on the mobility patterns of a group of people can be used for designing new and more efficient public transportation systems.

Despite the advantages that this information can provide –for example, to advise individuals for specific routes to avoid traffic jams–, it is clear that the collection and storage of such data raises important ethical issues, such as those concerned with the information security and users’ privacy. It is essential that the storage and processing of this information is carried out in a way that ensures the privacy of individuals whose data are collected or who want to enjoy the benefits of this technology.

The project involves designing systems that allow data collection with the required degree of privacy through the use of specific cryptographic protocols, combined with data mining and managing large amounts of data (big data).

Blockchain, and more broadly Distributed Ledger Technology (DLT), extends far beyond cryptocurrencies, revolutionizing various industries and enabling innovative business models based on decentralized services. Blockchain enhances security, privacy and transparency, eliminates intermediaries, empowers end-users, and facilitates new use cases previously deemed unfeasible. Current blockchain projects span numerous domains, including cryptocurrencies, payment systems, supply chains, e-health, e-voting, decentralized identity, and the collaborative economy.

Key Research Focus Areas:

- Enhancing Current Blockchain Technology: Improving security, privacy, scalability, interoperability and overall efficiency.

- Exploring Second Layer Protocols: Investigating advanced protocols to augment blockchain performance.

- Innovative Decentralized Services: Proposing new services or protocols where blockchain is a fundamental enabler.

- Blockchain and Artificial Intelligence Integration: Exploring the synergy between blockchain technology and AI.

- Blockchain Techniques for IoT security: Researching and evaluating efficient techniques/architectures for decentralization of IoT security.

This research line is not confined to the topics listed above. We actively welcome and encourage other related research proposals that align with our overarching goals of advancing blockchain technology and its applications.

The thesis is focused on the proposal to create a "Digital Chain of Custody" to ensure that the digital evidence (information or data, stored or trans-mitted in binary form which has been determined, through the process of analysis, to be relevant to the investigation) will be accepted in international court proceedings, so in it will be guaranteed the principles of identification, preservation, securing and posterior analysis.

After to establish a clear procedure, it will proceed with the second part: to create an artifice which it is able to comply with the procedure and it should take into account a set of items as: digital evidence acquisition and metadata associated (video, audio, photographs or files in general), probe localization, timestamp and secure communication capabilities. This device will be the starting point of the "Chain".

A new study of methods and applications in order to detect tampered multimedia content. Using Machine Learning and Artificial Intelligence techniques, the final method and application will be detect all modification media content, sound, video or images. Using techniques of steganography and steganalysis.

Data mining technologies have been constantly improving from last 20 years, the increasing computational power and storage capacity have allowed impressive accomplishments on the Artificial Intelligence and Machine Learning algorithms. 

This progress has been powered by the data collection through pervasive sensing by the Internet of Things and of smart-devices (such as smart-watches, smart-meters, etc.). As users’ data is collected in real-time, this must be carried out in a privacy-preserving manner not only to fulfill legal and ethical requirements but also individuals’ expectations. A user-centered (or local) approach for privacy protection may increase users’ confidence, through transparency and control. 

The aim of this proposal is to develop user-centered technologies for privacy protection of time-series obtained from sensors (such as location, health, behavioral or relational data).

We will study the guarantees provided by aggregation and randomized response methods to attain Local Differential Privacy. We will apply them to protect data that may be used for recommender systems, sequential pattern mining, complex networks analysis, predictions and decision making. 

The main contributions of this project will be to provide local algorithms for data protection and to analyze and develop strong guarantees of privacy for dynamic data.

Some relevant of differentialy private technologies, are google’s RAPPOR (Randomized Aggregatable Privacy-Preserving Ordinal Response) [1] or the US Census Bureau product called OnTheMap [2].

REFERENCES

[1] U. Erlingsson, V. Pihur, and A. Korolova. Rappor: Randomized aggregatable privacy-preserving ordinal response. In CCS, 2014.

[2] A. Machanavajjhala, D. Kifer, J. Abowd, J. Gehrke, and L. Vilhuber. Privacy: Theory meets practice on the map. In ICDE, 2008.

This research thesis proposes a novel framework leveraging Artificial Intelligence (AI) techniques, specifically machine learning (ML) and deep learning (DL) algorithms, to effectively distinguish between malware-infected files and clean files. The primary aim is to develop a proactive method for detecting intrusions, malware, and ransomware before attackers gain control over information systems. This approach addresses the limitations of traditional signature-based detection methods by providing a more adaptable and accurate solution to evolving cyber threats.

The framework builds upon existing research in the field, such as the works of Hussain et al. (2022) and Gavriluţ et al. (2009), which explored the application of ML algorithms for malware detection. By enhancing these methodologies, the proposed research will incorporate advanced techniques, such as neural networks and ensemble learning, to improve detection rates and reduce false positives and negatives. The adaptability of these AI models is crucial for identifying zero-day attacks and polymorphic malware, which often evade conventional detection systems.

Related work:

* Hussain, A., Asif, M., Ahmad, M.B., Mahmood, T., Raza, M.A. (2022). Malware Detection Using Machine Learning Algorithms for Windows Platform. In: Ullah, A., Anwar, S., Rocha, Á., Gill, S. (eds) Proceedings of International Conference on Information Technology and Applications. Lecture Notes in Networks and Systems, vol 350. Springer, Singapore. doi.org/10.1007/978-981-16-7618-5_53

* D. Gavriluţ, M. Cimpoeşu, D. Anton and L. Ciortuz, ""Malware detection using machine learning,"" 2009 International Multiconference on Computer Science and Information Technology, 2009, pp. 735-741,  doi: 10.1109/IMCSIT.2009.5352759.

Edge computing decentralizes data processing, bringing computing resources closer to where data is generated, which significantly reduces latency, enhances privacy, and increases efficiency. However, the proximity of edge nodes to devices, combined with their widespread distribution, creates a larger and more complex attack surface that traditional cybersecurity measures often struggle to protect. This demands innovative, robust security solutions that can adapt to the unique requirements of edge-based environments.

This research line will explore technologies like federated learning (FL), which enables collaborative model training across distributed devices without sharing raw data, preserving privacy and reducing network strain. Additionally, graph neural networks (GNNs) will be leveraged for their ability to model relationships and dependencies within complex, interconnected edge networks, aiding in anomaly detection and efficient resource allocation. Another key area of focus is zero-trust architectures, which verify each transaction or connection independently, enhancing security at each node. Together, these technologies aim to create a resilient cybersecurity framework for edge computing environments.

Key application areas for this research line include cyber-physical systems (CPS) and 6G networks. CPSs integrate cyber and physical components, enabled by modern sensor, computing, and network technologies. These systems capture data from physical objects and transmit it through networks to a central control system. Architectures incorporating edge, fog, and cloud computing then process this data, enabling real-time decisions that drive actions in the physical environment. In 6G networks, edge computing supports ultra-low latency and advanced applications like augmented reality, but the distributed nature of nodes increases vulnerabilities. This research line focuses on creating security solutions that protect both data and device integrity in these critical edge environments.

AI models, widely used in domains such as finance, healthcare, autonomous systems, and smart infrastructure, face growing cybersecurity and privacy risks. These models are vulnerable to adversarial attacks, data poisoning, model inversion, and membership inference attacks, each of which can expose sensitive data or compromise model integrity. Such threats can distort model predictions, breach user privacy, and undermine trust in AI applications.

This research line focuses on identifying and mitigating the cybersecurity and privacy threats unique to AI models. Areas of exploration include developing robust defenses against adversarial attacks, implementing privacy-preserving techniques like differential privacy and federated learning, and establishing secure model lifecycle practices. Additionally, the use of multi-party computation and homomorphic encryption will be explored as methods to enhance privacy, allowing multiple parties to perform computations on encrypted data without revealing the underlying information. By addressing these challenges, this research aims to enhance the security, privacy, and reliability of AI models across critical applications.

As data collection and sharing have become central to research, policy-making, and business practices, the ability to protect sensitive information while still providing useful insights is crucial. 

Data privacy is not just an ethical obligation but a practical necessity. Earning the trust of data subjects, for instance, enables us to gather more accurate information. Moreover, data protection regulations, such as the GDPR, place data privacy at the forefront of any personal data processing activity, particularly when sensitive data or data about vulnerable individuals is being processed.

Statistical Disclosure Control (SDC) is a set of techniques that aim to minimize the risk of disclosing sensitive information while preserving, as much as possible, the statistical value of the data.

Traditionally, SDC techniques alter the data in specific ways to prevent situations that are known to be disclosive. For instance, by generalizing or suppressing sets of attributes that could lead to the re-identification of records. While still very relevant, traditional techniques lack formal privacy guarantees. This has led to the development of novel approaches by the computer science community. The most prominent of these is differential privacy, which offers privacy regardless of the external information available but, on the other hand, has a significant impact on data utility.

The popularity of differential privacy, together with its impact on data utility, has led to applications where achieving strong privacy guarantees may be challenging.

While formal privacy models may be considered a central topic in the SDC community, context is critical in SDC, and there is no one-size-fits-all solution. Therefore, this line of research remains open to both traditional and formal privacy approaches.

The main objective of this line of research lies in the development of data mining tools for knowledge discovery from large amounts of data, represented in multidimensional spaces and coming from different kinds of sources: e.g. design and implementation of parameter-free hierarchical clustering algorithms, which do not require any prior knowledge about the data, including the number of clusters to identify.

The scope of application of these tools is broad and varied, and the data can be found in a wide variety of research fields, including economics, psychology, climatology, biology, computational genomics, statistics, education, information retrieval, text mining, computer vision and machine learning, among many others.

Online higher education introduces a way of transcending formal higher education by realizing technology-enhanced formats of learning and instruction and by granting access to an audience beyond students enrolled in a conventional higher education institution. However, although online higher education has been reported as an efficient and important educational context, there are a number of issues and problems related to the educational aspect. More specifically, there are higher dropout rates during a course, little participation and a lack of student motivation and engagement overall. This may be due to one-size-fits-all instructional approaches and very limited commitment to student-student and teacher-student collaboration.

This thesis aims to enhance the online higher Education experience by integrating:

• collaborative settings based on conversational agents (CAs) both in synchronous and asynchronous collaboration conditions; and
• screening methods based on learning analytics (LA) to support both students and teachers during an online higher education course.

CAs guide and support student dialogue using natural language both in individual and collaborative settings. Moreover, LA techniques can support teachers' orchestration and students' learning during online higher education by evaluating students' interaction and participation. Integrating CAs and LA into online higher education can both trigger peer interaction in discussion groups and considerably increase the engagement and commitment of online students (and, consequently, lower dropout rates).

Nowadays, many systems help students to learn. Some of them aid students in finding learning resources or recommending exercises. Others aim to help the student in the assessment phase by giving feedback. Furthermore, others monitor the student's progress during the instructional process to recommend the best learning path to succeed in the course. Depending on the objectives/competencies of the subject, some features are more suitable than others.

This research line proposes to work in intelligent learning systems based on artificial intelligence techniques focusing on the following topics:

• Predictive analytics based on machine learning algorithms

• Early warning systems able to detect at-risk students 

• Automatic feedback and nudging based on generative artificial intelligence

• Ethical issues (fairness, transparency and explainability)

• Data visualization and dashboards

• Gamification

• Virtual educational advisor (chatbots)

During the last years the board game field have experimented a great expansion in the means of the quantity of boardgames available, if the variety of them, of the broad coverage of topics they address and the variety of mechanics they provide. They have great potential to become a great tool for learning, as many research studies show. 

In this research line, we would like to address the latest innovations of using boardgames for learning and to explore the potential of using boardgames in the eLearning context and the mechanisms that appear in them.

Generative AI (GenAI) tools based on Large Language Models (LLMs) have demonstrated impressive performance in myriad types of programming tasks. As a result, their impact on introductory programming (CS1) courses should be studied in depth. This research line aims to explore the present realities and the future possibilities in how Generative AI is impacting, and may further impact on introductory programming courses, including learning goals/outcomes, assessment, emerging pedagogies, and educational resources. Some research questions may be:

* When and how should GenIA tools be introduced in introductory courses?

* What types of impact GenIA-based activities or tools have on the CS1 students (i.e. performance, behaviour, understaning, etc.)? 

* What kinds of GenAI-based assignments could CS1 students do? For example, Kerslake et al. (2024) proposed two activities: (1) the first one involved students solving computational tasks by writing prompts to generate code; (2) the second one involved showing students a code fragment and asking them to demonstrate their understanding of the code by crafting a prompt that generates equivalent code.

* How can GenAI tools support CS1 students? (e.g. instant feedback, high-level problem solving advice, automated assessment, etc.)

* How can GenAI tools support CS1 instructors? (e.g. writing feedback, creation of assignments, etc.)

* Might pair programming evolve from two students working together into ""me and my AI""?

Many factors have been cited for poor performance of students in programming courses, especially in CS1. One of them is the assessment process.  The goal of this research line is to explore different aspects related to assessment in programming courses. Some research questions may be:

* How may assessment mechanisms impact on students' performance?

* How to design exams that really assess students' skills and knowledge? This includes to analyze aspects such as the delivery mode (f2f or online), the format (paper, computer, IDE), the duration, and so on.

* What impact do the different assessment strategies (e.g. contract grading, mastery-learning, second-change testing, etc.) in students' performance and satisfaction?

* What instruments are best for assessing CS1 students? (e.g. rubrics, tests, etc.)

* What aspects should be evaluated in programming courses? (e.g. output/behaviour, code quality, etc.) How should they be evaluated?

In addition to the previous questions, the design and development of tools that support the assessment process in programming courses, such as automated grading (i.e. online judges or automatic exam generators) are welcomed as well.

Using humanistic HCI methodologies, with the possibility to prototype, develop or analyse real data from a growing network of students using Folio (folio.uoc.edu) and Agoras, this line studies autonomy, identity and collective learning in open, ethical and participatory digital ecosystems bridging institutional (LMS) and personal (CMS/WordPress) spaces.

Model-driven development is a software development approach that attempts to reduce development costs by focusing on producing software models (usually specified by UML) rather than code, and relying on tools to automatically generate the final implementation from these models.

This line of research will investigate techniques and tools to support model-driven software development processes (model transformations, executable models, domain specific languages). The focus of this work will be on developer productivity improvements and the quality of the final software product.

Software analytics is the study of all data related to software and its engineering processes in order to better understand how software is built. The goal is to be able to predict and improve important quality factors of software artifacts. Software analytics includes the analysis of the program code but we are interested also in the analysis of all the collaboration and social aspects around it (who is the community that builds the software? How are they organized? What best practices do they follow? Etc.).

Bugs in software systems may lead to catastrophic consequences, especially in safety-critical systems such as medical or aerospace software. Testing and code reviews can reduce the defect rate, but sometimes a higher level of assurance is required. To this end, formal methods are a family of techniques that analyse a mathematical description of the system in order to ensure its correctness.

Some techniques used in the formal verification of software are model checking, theorem proving and static analysis. A problem shared by these approaches is their high computational complexity, which can limit their applicability in real-world examples. This line of research will consider pragmatic approaches for ensuring the quality of software systems at an industrial scale, considering key issues such as usability, efficiency and applicability.

This research line will study both the impact of AI in software engineering and the application of software engineering techniques in the development of AI applications.

AI in software engineering will study how AI techniques can improve and speed up software development. For instance, by recommending improvements to the software models based on past projects or general language models. Software engineering for AI will focus on defining new languages and frameworks to facilitate the creation of intelligent software, especially targeting hybrid applications where the system needs to combine smart components (e.g. a chatbot) with a “traditional” ones (e.g. a database that contains the answers to the chatbot questions).

Automated ubiquitous sensing will represent a major change in current societal challenges in terms of efficiency (i.e. automatization), sustainability (better models and decision-making), and social wellbeing (security and improved job market). Automated ubiquitous sensing would, for instance, detect humidity or thin water ponds on a road avoiding possible car accidents in a smart city context, improve recycling processes thanks to automated items classification, or predict dehydration of a person in an ambient assisted living environment.

Current solutions for automated sensing, including active and passive sensors, use specific circuitry for sensing besides wireless communications technologies to transmit the measurements. This approach implies battery management complexity and/or expensive customized technology, being inappropriate from the sustainability perspective.

This proposal aims to exploit backscatter low-cost communication technologies like Ultra High Frequency (UHF) Radio Frequency Identification (RFID) to act as sustainable ubiquitous sensors. The goal is to improve the integration of passive sensing in the environment, towards a seamless sustainable digitization. For more information see [1,2,3,4].

[1] Floerkemeier, Bhattacharyya and Sarma (2010): https://doi.org/10.1007/978-1-4419-1674-7_21

[2] Melià-Seguí and Vilajosana (2019): https://doi.org/10.1109/RFID.2019.8719092

[3] Lejarreta-Andrés, Melià-Seguí, Bhattacharyya, Vilajosana and Sarma (2022): https://doi.org/10.1109/JSEN.2022.3188936

[4] Melià-Seguí, Bhattacharyya, López-Soriano, Vilajosana and Sarma (2024): https://doi.org/10.1109/JSEN.2023.3339117

Smart cities are a new scenario where a variety of data sources open the door to innovative applications, with the final goal of improving the quality of life of citizens, industry competitiveness and government. The Internet of everything is responsible for collecting and transmitting this data at different levels, by using a wide variety of sensors (including peoples' smartphones). We propose to make use of contextual intelligence tools and techniques (like data mining or statistical learning) to process signals (accelerometers, acoustic, etc.) with the overall goal of transforming the above collected data into useful information, enabling new applications or improving existing ones within the ambient intelligence context.

The demand for highly flexible and adaptable communications networks, especially cellular systems, has been a core driver over the last decade. This has spurred academic and industry efforts to design architectures that enable fully autonomous management and orchestration, largely by exploiting network data with Generative AI solutions [1, 2].
In parallel, the Intent-Based Networking (IBN) paradigm has gained traction, driven by the inherent complexity of modern networks and the need to manage diverse service Quality of Service (QoS) requirements. IBN harnesses the power of Large Language Models (LLMs) to streamline the configuration of complex networks and improve the comprehension of autonomous network decisions [3, 4, 5].
Moving toward real autonomous network management, Agentic AI native solutions are now emerging as flexible, scalable, and reliable frameworks for the future 6G network infrastructure [6]. Despite the clear research objective, current solutions are in their infancy, underscoring the immediate need for novel contributions.
This research proposal directly addresses this gap by contributing to the definition and development of Agentic AI-based network management and orchestration frameworks. Our work will integrate forefront AI and networking technologies, ensuring alignment with key network architectures and close coordination with standardization bodies.

References:

[1] L. Bariah and M. Debbah, "AI Embodiment Through 6G: Shaping the Future of AGI," in IEEE Wireless Communications, vol. 31, no. 5, pp. 174-181, October 2024, doi: 10.1109/MWC.015.2300521.
[2] Y. Xiao, G. Shi and P. Zhang, "Toward Agentic AI Networking in 6G: A Generative Foundation Model-as-Agent Approach," in IEEE Communications Magazine, vol. 63, no. 9, pp. 68-74, September 2025, doi: 10.1109/MCOM.001.2500005.
[3] Y. Wang, C. Yang, T. Li, Y. Ouyang, X. Mi and Y. Song, "A Survey on Intent-Driven End-to-End 6G Mobile Communication System," in IEEE Communications Surveys & Tutorials, doi: 10.1109/COMST.2025.3575041.
[4] A. Mekrache, A. Ksentini and C. Verikoukis, "Intent-Based Management of Next-Generation Networks: an LLM-Centric Approach," in IEEE Network, vol. 38, no. 5, pp. 29-36, Sept. 2024, doi: 10.1109/MNET.2024.3420120.
[5] A. Mekrache, A. Ksentini and C. Verikoukis, "DMO-GPT: An Intent-Driven Framework for Distributed 6G Management and Orchestration," in IEEE Communications Magazine, doi: 10.1109/MCOM.001.2500258.
[6] M. Elkael, S. D'Oro, L. Bonati, M. Polese, Y. Lee, K. Furueda and T. Melodia, “AgentRAN: An Agentic AI Architecture for Autonomous Control of Open 6G Networks”, Arxiv.org, Aug. 2025. https://arxiv.org/abs/2508.17778

GNSS underpins safety-critical navigation for UAVs, road vehicles and rail, yet urban canyons, unintentional interference, intentional jamming and spoofing can degrade or corrupt positioning. We propose a research programme that (i) protects the signal chain against denial and deception, and (ii) reconstructs or recovers the navigation state so that accurate positioning is maintained, even under severe interference.

The Ambient IoT paradigm envisions a pervasive network of battery-free or energy-autonomous devices, embedded seamlessly in the environment, capable of sensing, communicating and interacting without explicit human intervention. Despite recent progress, major challenges still remain in realizing robust, large-scale Ambient IoT deployments and ensuring their harmonious coexistence with legacy networks.
This project aims to develop solutions for boosting the link budget of passive (and semi-passive) devices to increase the detection range, transmit larger volumes of data and achieve higher robustness. To this end, the project will investigate novel reconfigurable backscatter mechanisms and cooperative relaying schemes to enhance the communication range and reliability under stringent power constraints. It will explore the use of energy from nearby devices, base stations and environmental sources (i.e. solar) to design efficient methods and protocols for energy harvesting.
The project is a research collaboration with the Aalborg University (AAU), Denmark.

[REF] Perovskite PV-Powered RFID: Enabling Low-Cost Self-Powered IoT Sensors

The Ambient Internet of Things (AIoT) relies fundamentally on pervasive, zero-power sensing devices (truly battery-free or passive), seamlessly embedded in the environment for autonomous data collection. However, achieving robust, wider-area truly passive AIoT deployments is constrained by critical design trade-offs, particularly antenna design and substrate materials for performance optimization in different environments.

This research line aims to develop solutions for boosting the link budget of passive (and semi-passive) devices, focusing on improvement of the end device. We aim to explore different technologies and techniques, including chip-enabled solutions (i.e. reconfigurable backscatter mechanisms) as well as chipless approaches to ensure robustness and improved-range operation under stringent power constraints, enabling the next-generation AIoT devices. This project is included within the EU-funded Pathfinder GAIA project (2026-2028) [1].

[1] https://www.uoc.edu/en/news/2025/gaia-project-for-sustainable-iot

Human-computer interaction (HCI) is organized around three main elements: design, technology and people. It mainly focuses on the definition, design and evaluation of interactive products, tools and systems from a user-centered perspective. Interaction design researchers explore, design, build and test interactive products and systems with user-experience goals in mind. HCI provides a set of models, methods and techniques to user centered interaction design research that take into account the design process and key activities such as user research, ideation, interaction definition, prototyping and user testing. This line explores research around interaction design processes and methods, interaction co-design, conversational interfaces, gesture interfaces or tangible interaction.

This research proposal is focused on the relationship between HCI (Human-Computer Interaction) and TEL (Technology-Enhanced Learning): how design and technology impact students and their learning. Interaction design and human-centered design can transform learning experiences and provide e-learning environments, learning resources and educational tools that take into account learners’ needs, wants and limitations. Therefore, this research proposal focuses on design, technology and learning in order to provide interactive and user-centered technology-enhanced learning. 

This research topic includes:

•  Learning Toolkits. The Design Toolkit (design-toolkit.uoc.edu, 2017) is an open educational repository presented as a toolkit -a set of tools for designers to learn and practice. This research focuses on design toolkits, aiming to define, assess and optimize “expert workflows”  in order to address the needs of both student and professional (expert) designers.
•  Interactive learning tools. Folio.uoc.edu (2018) is a social network learning platform that integrates Learning Management Systems with Content Management Systems, creating a tecno-pedagogical model based on self-identification, social learning and professional development. This research line delves into the use of personal e-portafolios in rich networked environments with a focus on the implications and potentialities of its interfaces and interactions.
•  Interactive learning tools. CodeLab.uoc.edu (2020) is an ongoing project that provides a laboratory-based learning tool for non-STEM students to learn to code through a hands-on approach. Students learn by doing, following learning itinerary, exploring how to solve problems, completing activities and engaging in discussions with peers and/or teachers to address challenges.
•  Embodied interaction in technology enhanced learning. How to design embodied interactive learning experiences in the context of e-learning and asynchronous and distance learning?

Augmented and Virtual Reality are more and more present in our personal and social lives. They define a new media, requiring real-time technologies where the interaction is essential, but also where the interface plays a key role, allowing 3D interaction in Immersive Virtual Environments. 

This research aims to focus not only on the technological aspects, but also on other aspects such as usability, ethical issues, or human factors. It intends to encompass all areas related to virtual reality (VR), including augmented reality (AR), mixed reality (MR), and 3D user interfaces (3DUIs).

Moreover, it focuses on advances and novelties in theories and methods but also on applying existing ideas to solve interesting problems with an innovative approach or on the experience learned from developing effective systems integrating such techniques and technologies.

Keywords: 3D User Interfaces, Augmented Reality, Avatars, Diversity and gender issues, Embodied agents, virtual humans and (self-)avatars, Ethical issues, Human factors and ergonomics, Immersive / 360° video, Immersive games, Immersive Virtual Environments, Perception and cognition, Presence, body ownership, and agency, User experience and usability, Virtual Reality. 

Digital making’ is a keyword that is used regularly in discussions about technology education and it has a twofold meaning: “making digital things” and “making things using digital tools”.

Although the term is descriptive, creating artifacts with digital technology, the focus of its use encompasses everything from creating software and hardware using, through to designing and constructing physical objects using digital fabrication tools – CNC machines, 3D printers and laser cutters.

This research proposal pays special attention to the relationship between digital making and technical education for no STEM students, exploring the potential of digital making tools in order to nurture technical knowledge for digital creation such as creative coding, generative art and design and digital fabrication.

This research proposal focuses on critical interaction design in order to explore functional, non-functional, aesthetic, cultural, speculative and political dimensions of interfaces. 

Speculative, critical and discursive design constitute the methodological approaches to explore critical interaction design. The research in this area includes discursive, contextual and analytical research and also prototyping, development and building artifacts as means to test ideas and validate hypotheses.

This research topic includes interactive media archaeology; interaction for social good and justice vs. dark patterns and surveillance capitalism; measuring agency on interactive artifacts and/or conversational interfaces; user-centered interaction design in interactive media art.

La proposta de tesi s'emmarca en l'àmbit del digital making i cerca generar investigacions finalitzades al disseny i fabricació de materials educatius de codi obert mitjançant l'ús de tecnologies CAD/CAM.

Les investigacions s'iniciaran en col·laboració amb el professorat de centres educatius i es duran a terme amb dinàmiques de co-disseny. El codi font dels productes i tota la informació necessària a la fabricació estaran publicats en obert i podran ser descarregats per a la seva fabricació per qui hi estigui interessat.

Realistic generation of talking faces for natural interaction. Depending on the interests of the doctoral student, the image or audio part can be developed to different degrees.

Data visualization is a multidisciplinary field that impacts a wide range of domains. Every data visualization is inherently subjective, embodying a specific perspective of the world; it presents reality in a particular way by highlighting certain aspects of the information while omitting or concealing others. This doctoral thesis proposal considers both the study of visualizations and their interaction interfaces, as well as the creation of critical and speculative visualizations. Developed through a methodology rooted in hands-on practice and reflective analysis, and adopting a critical, situated, and contextual theoretical approach, the theses in this proposal may focus on the following topics:

•  Design of critical visualizations in educational contexts.
•  Development of analytical frameworks for visualization interfaces.
•  Analysis and creation of artistic and speculative visualizations that question dominant socioeconomic models.
•  Study and design of data visualization for generative AI processes.
•  Visualization as a methodology for speculative research.

Video game platform studies research how platforms, both hardware and software, shape game development and player experiences, covering a range of topics, spanning platform-specific technical limitations to the cultural and social impact of platforms. In that regard, Japanese video games have unique characteristics that differentiate them from other international productions, such as their gameplay and their graphic and narrative style, which can sometimes be limited by technical and design decisions. To this end, developers create specific graphics engines and tools to assist them in the development of their games. This line of research aims to analyse and understand the technical, creative and strategic factors that Japanese video game companies take into account to overcome the limitations of the specific hardware on which their video games will run, while maintaining their creative essence.

Almost any action occurs in a place and in a moment in time. And this fact has taken on the utmost importance in recent years with the spread of mobile devices capable of providing GPS location and associated services to the point where we are. However, although these systems have become usual nowadays, there are still many possibilities to explore and even, situations in which those systems don’t work, like indoor spaces.

In this area of research we seek answers to questions like: What are the new possibilities that Galileo will open up? What role will the user play within these location-based applications? What happens in indoor environments? Which possibilities do the deployment of 5G network open to location systems?

The area is focused on the advancement of location-based systems (LBS) and context aware recommender systems (CARS) in outdoor and indoor environments, working on mobile devices. Basically its goals are:

a) To increase the functionalities of LBS and CARS to make them work in indoor environments; and to focus them on specific sectors like health.

b) To include personalization in the LBS, in addition to time and position, in order to improve and optimize users’ experiences. In this approach, semantic web technologies are currently used.

In recent years, there has been an increase in the number of outbreaks of infectious diseases that had previously been categorized as being under control. HIV and tuberculosis are between these diseases and, most recently, the covid-19 outbreak. 

On the other hand, the increase of exotic travels where diseases like malaria are still present, has bring it to countries where it has been banished.  How to help health authorities to prevent the spread of these diseases?

Since most of population uses a smartphone, and they have location capabilities, preventive and informative actions through smartphone, based on location and context, has proven some positive effects.

This proposal is focused on using apps and all the available possibilities that the sensors in a smartphone offer to prevent the spread of infectious diseases and to promote healthy habits.

Urban areas are complex systems composed of interconnected networks of infrastructure, people, and their activities. Understanding the structure and dynamics of these systems is crucial for addressing urban challenges and promoting sustainable development, combining both data- and theory-driven developments. This research line focuses on applying complex systems science to analyze urban areas, with a particular emphasis on transportation systems, pedestrian dynamics, and the impact of urbanization on the environment.

Research Focus
This research line investigates the following key areas:
- Sidewalk networks and pedestrian dynamics [1-5]: Analyzing the structure and usage patterns of pedestrian infrastructure to understand pedestrian movement, optimize walkability, and related issues like pollution exposure.
- Dynamics of transportation systems [6, 7]: Studying emergent behaviors of urban transportation systems, such as congestion and disruptions, including multimodal transportation networks.
- City resilience [4, 8, 9]: Assessing the ability of urban systems to withstand and recover from shocks and stresses, such as natural disasters, economic downturns, and pandemics.

Research Methods
This research line employs a variety of methods, including:
- Network analysis: Representing urban systems as networks provides a simple and elegant approach to model the intricate relationships between various urban components, such as transportation infrastructure, social connections, and economic activity.
- Agent-based modeling: Simulating the behavior of individual agents (e.g., pedestrians, vehicles) to understand emergent patterns and dynamics at the system level.
- Data analysis and Geographical Information Systems (GIS): Utilizing large spatial datasets from various sources, such as sensors, GPS traces, and social media, to understand urban patterns and urban phenomena in a geographical context.
- Computer vision: Employing image processing and machine learning techniques to extract information from urban imagery, such as pedestrian detection, sidewalk segmentation, and streetscape analysis.

[1] J Mateu Armengol et al. City-scale assessment of pedestrian exposure to air pollution: A case study in Barcelona. Urban Climate 58 (2024)
[2] D Rhoads et al. The inclusive 15-minute city: Walkability analysis with sidewalk networks. Computers, Environment and Urban Systems, 100, 101936 (2023).
[3] D Rhoads, C Rames, A Solé-Ribalta, MC González, M Szell, and J Borge-Holthoefer. Sidewalk networks: review and outlook. Computers, Environment and Urban Systems, 106, 102031 (2023).
[4] D Rhoads, A Solé-Ribalta, MC González, J Borge-Holthoefer. A sustainable strategy for Open Streets in (post) pandemic cities. Communications Physics 4 (1), 1-12 (2021)
[5] C Bustos, et al. Explainable, automated urban interventions to improve pedestrian and vehicle safety. Transportation Research Part C: Emerging Technologies 125, 103018 (2021)
[6] A Lampo, J Borge-Holthoefer, S Gómez, A Solé-Ribalta. Multiple abrupt phase transitions in urban transport congestion. Physical Review Research 3 (1), 013267 (2021)
[7] A Solé-Ribalta, S Gómez and A Arenas. Congestion induced by the structure of multiplex networks. Physical Review Letters 116(10), 108701 (2016)

[8] C Li, et al. From flood-prone to flood-ready: The restoration-adaptation interplay in building resilient multimodal transport networks. Reliability Engineering & System Safety, 111737 (2025)
[9] C Li, et al. Adaptive capacity for multimodal transport network resilience to extreme floods. Nature Sustainability, 1-12 (2025)

This research line delves into the intricate dynamics of online communication and collaboration, with a focus on understanding how individuals and groups interact and organize in the digital age. We employ computational techniques to analyze large-scale social phenomena and uncover the patterns that shape online behavior.

Research focus:
- Collaboration patterns in groups –from scientific communities to international organizations [1-3]: We investigate how agents and entities, at different aggregation levels, share knowledge and interact. This includes studying the formation of research teams,
cooperation in software repositories, or cultural and diplomatic ties in historical institutions.
- Structural flexibility and synchronization in online communication systems [3-5]: Online communication platforms, like social media networks, are constantly evolving. We study how these systems react/adapt to external events and internal pressures, and how
their structure affects the flow of information and the formation of consensus.

Research methods:
- Network analysis: We use network theory and algorithms to study the relationships and interactions between individuals and groups in off- and on-line environments.
- Computational modeling: We develop computational models to simulate and study the dynamics of group formation and evolution. This includes the use of agent-based modeling to explore the emergence of complex social phenomena from simple mechanisms.

[1] R Rodríguez-Casañ, et al. Analysing inter-state communication dynamics and roles in the networks of the International Institute of Intellectual Cooperation. Humanities and Social Sciences Communications 11: 1408 (2024)
[2] R Rodríguez-Casañ, MJ Palazzi, A Solé-Ribalta, M Nordberg, A Canals, J Borge-Holthoefer. The robust-fragile duality of the ATLAS collaboration network. EPJ Data Science 14 (1), 57 (2025)
[3] MJ Palazzi, J Cabot, JLC Izquierdo, A Solé-Ribalta, J Borge-Holthoefer. Online division of labour: emergent structures in Open Source Software. Scientific Reports 9, 13890 (2019)
[4] MJ Palazzi, A Solé-Ribalta, V Calleja-Solanas, S Meloni, CA Plata, S Suweis, J Borge-Holthoefer. An ecological approach to structural flexibility in online communication systems. Nature Communications 12 (1), 1-11 (2021)
[5] J Borge-Holthoefer, RA Baños, C Gracia-Lázaro and Y Moreno. The nested assembly of collective attention in online social systems. Scientific Reports (2017)
[6] J Borge-Holthoefer, N Perra, B Gonçalves, S González-Bailón, A Arenas, Y Moreno and A Vespignani. The dynamic of information-driven coordination phenomena: a transfer entropy analysis. Science Advances 2(4), e1501158 (2016)

This research line explores the complexity of mutualistic interactions, focusing not only on the classical ecological applications –mainly plant-pollinator networks–, but also in connection with other socio-urban systems. Our research spans from the development of novel data collection
methods to the analysis of network patterns and their implications for community stability and persistence.

Research focus:
- Developing and implementing automated monitoring systems [1]: We design and deploy automatic camera systems equipped with deep learning algorithms to monitor plant-pollinator interactions in the field. This technology allows us to capture both diurnal
and nocturnal interactions, providing a comprehensive view of pollination dynamics.
- Analyzing the structure of mutualistic networks [2-4]: We investigate the emergent patterns in natural (e.g., plants and pollinators) and social (e.g., users and hashtags) ecosystems, focusing on nestedness, modularity, and in-block nestedness. We explore
how these patterns affect, for example, the stability and feasibility of mutualistic communities.
- Understanding the dynamics of mutualistic interactions [2,3]: We study how plant-pollinator networks change over time, and how these changes affect the persistence of species and the functioning of ecosystems. We use computational models to simulate
the dynamics of these networks and explore the impact of environmental changes.
- Developing new methods for network analysis [5-7]: We develop new analytical and computational tools to study the structure and dynamics of mutualistic networks, e.g. methods for bipartite network randomization and pattern detection.

Research methods:
- Field experiments: We conduct field experiments to collect data on plant-pollinator interactions, using automated monitoring systems combined with computer vision techniques. Analysis of images and videos of plant-pollinator interactions, species
identification, movement tracking.
- Network analysis: We employ network theory and algorithms to study the patterns of connections between species, with special attention to bipartite networks’ specificities.
- Computational modeling: We develop computational models to simulate the dynamics of plant-pollinator networks and explore the impact of environmental changes.

[1] PE Serra‐Marin, et al. Comparative assessment of automated and manual monitoring in comprehensive plant–pollinator communities. Methods in Ecology and Evolution (2025)
[2] A Lampo, MJ Palazzi, J Borge-Holthoefer, A Solé-Ribalta. Structural dynamics of plant–pollinator mutualistic networks. PNAS Nexus 3 (6) (2024)
[3] MJ Palazzi, A Solé-Ribalta, V Calleja-Solanas, S Meloni, CA Plata, S Suweis, J Borge-Holthoefer. An ecological approach to structural flexibility in online communication systems. Nature Communications 12 (1), 1-11 (2021)
[4] MJ Palazzi, J Cabot, JLC Izquierdo, A Solé-Ribalta, J Borge-Holthoefer. Online division of labour: emergent structures in Open Source Software. Scientific Reports 9, 13890 (2019)
[5] A Solé-Ribalta, CJ Tessone, MS Mariani, J Borge-Holthoefer. Revealing in-block nestedness: detection and benchmarking. Physical Review E 97 (6), 062302 (2018)
[6] MJ Palazzi, J Borge-Holthoefer, CJ Tessone, A Solé-Ribalta. Macro-and mesoscale pattern interdependencies in complex networks. Journal of the Royal Society Interface 16 (159), 20190553 (2019)

[7] MS Mariani, MJ Palazzi, A Solé-Ribalta, J Borge-Holthoefer, CJ Tessone. Absence of a resolution limit in in-block nestedness. Communications in Nonlinear Science and Numerical Simulation 94 (2021)

Medical image processing is a key step in the diagnosis of a large number of diseases. Nowadays, we can acquire images of the inside and outside of our bodies using a large variety of devices (ultrasound, magnetic resonance, optic tomography, computed tomography, etc.). Afterward, the acquired images usually need to be denoised, corrected for inhomogeneities, segmented, registered, etc. in order to be able to get relevant information to aid the clinical decision using image-based biomarkers. 

On this research line, we would like to explore the latest image processing challenges and develop new image-based biomarkers that aid clinicians in their daily work. This work will be done in collaboration with world-wide recognised clinical institutions in Barcelona. 

Disasters have a significant and increasing impact all over the world. There is a growing concern about them, so Disaster Risk Reduction (DRR) is increasingly in international agenda. This thesis proposal sets up the scientific and technical basis for a significantly improved resilience to hazards (such as climate related hazards, earthquakes, pandemics, etc.) and their human and socioeconomic impacts related to Suppply Chain Management (SCM).

The proposal is based on three principles, inspired by UN Sendai Framework and related to UN 2030 Agenda for Sustainable Development:

1) Focus on prevention and resilience building oriented. 2) Inclusive “whole-of-society” approach, to involve non-traditional stakeholders not usually involved in DRR planning and decision making (such as households, SMEs, NGOs, etc.). 3) Data-driven approach, to integrate in DRR planning and decision-making diverse types of data (including small data, thick data, and big data) from a wide range of sources, and including reuse of data.

This thesis proposal will conduct research about data-based instruments for SCM's optimization and resilience related to disasters and disruptions.

This thesis proposal focuses on case studies about misinformation or disinformation, through the application of quantitative data analytics methods, including artificial intelligence models, to amounts of digital content such as: social media, mainstream media news and reports, Wikipedia entries, literature about historical events, open data and/or other open or public domain sources. This content may be created, updated, searched, retrieved, summarized or selected automatically, through algorithms including artificial intelligence models. This digital content may be created, updated, influenced and/or used by a wide range of actors: citizens, anonymous agents or activists, governments and public agencies, companies, international organizations, political parties, social organizations, etc.
Research methodologies for these case studies will usually include the advanced conceptualization of misinformation and disinformation events, so as to enhance the intensive application of quantitative methods to trace and analyse them through amounts of digital content and logs. These quantitative methods may be combined when suitable with qualitative methods.

•  Cardoso, G.; Sepúlveda, R.; Narciso, I. (2022). Whatsapp and audio misinformation during Covid-19 pandemic. El Profesional de la Información https://doi.org/10.3145/epi.2022.may.21 
•  Cobarsí-Morales, J. (2022). Controversial ‘Black Kegend’ concept as misinformation or disinformation related to history: where do we go from here in the 21st century information field?. In: Smits, M. Proceedings iConference 2022 – Information for a Better World: Shaping the Global Future.
•  Salaverría, R., & León, B. (2022). Misinformation beyond the media: ‘fake news’ in the big data ecosystem. In: Vázquez-Herrero J., Silva-Rodríguez A., Negreira-Rey M.C., Toural-Bran C., López-García X. (2022). Total Journalism. Models, Techniques and Challenges (pp. 109-121. Studies in Big Data, 97. Springer Nature. Cham: Springer. DOI:10.1007/978-3-030-88028-6_9
•  Meel, P.; Vishwakarma, D.K. (2019). Fake news, rumor, information pollution in social media and web: A contemporary survey of state of the arts, challenges and opportunities. Expert Systems With Applications https://doi.org/10.1016/j.eswa.2019.112986

Neuroaxonal anatomy and function are affected by multiple sclerosis (MS) disease which, in turn, impacts the brain structure, organization, and function. In general, networks representing particular brain aspects (morphology, structure, or dynamics) are studied independently to understand and predict individual brain damage effects [1]. Designing a single unified model to jointly study these multiple aspects is necessary to understand neurological diseases. 

Within this PhD we would like to introduce an interconnected multi-layer framework for the joint analysis of morphological, structural, and functional networks. Therefore, we aim to define a multi-layer scheme that allows us to combine the information of morphological, structural, and functional networks into a single scheme in order to better assess brain damage effects and evolution of MS patients. Then, it is very relevant to define or adapt graph-mining metrics to evaluate and quantify the deterioration of the connectivity of the brain using the new multi-layer scheme.

This work will be done in close collaboration with the Multiple Sclerosis group led by Dr. Sara Llufriu at the IDIBAPS-Hospital Clinic, a world-wide recognized clinical institution.

Ref: Jordi Casas-Roma, Eloy Martinez-Heras, Albert Solé-Ribalta, Elisabeth Solana, Elisabet Lopez-Soley, Francesc Vivó, Marcos Diaz-Hurtado, Salut Alba-Arbalat, Maria Sepulveda, Yolanda Blanco, Albert Saiz, Javier Borge-Holthoefer, Sara Llufriu, Ferran Prados; Applying multilayer analysis to morphological, structural, and functional brain networks to identify relevant dysfunction patterns. Network Neuroscience 2022; 6 (3): 916–933. doi: https://doi.org/10.1162/netn_a_00258

An integrative approach to predict disability in multiple sclerosis through image analysis, synthetic data generation and the provision of a federated learning platform. This PhD is going to be in collaboration with the research group ImaginEM at Hospital Clinic BCN.

The primary clinical objective of this PhD project is to develop robust predictive models for disability progression in individuals diagnosed with multiple sclerosis [1]. This clinical objective involves leveraging advanced imaging techniques, clinical data, and machine learning algorithms to identify early indicators of disability. The successful candidate will collect and analyse multimodal imaging data, including but not limited to MRI, CT scans, and patient clinical records. The aim is to enhance our understanding of disease progression patterns, ultimately contributing to developing personalised treatment plans.

In parallel with the clinical objectives, the PhD candidate will be responsible for developing an open federated learning system platform tailored to analyse medical images [2,3]. This platform will facilitate collaboration and data sharing across healthcare institutions while ensuring data privacy and security. The candidate will design and implement federated learning algorithms, enabling the aggregation of insights from diverse datasets without centralised data storage. Developing a synthetic data generation module [4] will also be crucial, providing a more extensive and diverse dataset for training machine learning models.

[1] https://jnnp.bmj.com/content/early/2023/07/19/jnnp-2022-330203.abstract

[2] https://www.worldscientific.com/doi/abs/10.1142/S0129065722500496

[3] https://link.springer.com/chapter/10.1007/978-3-031-22356-3_12

[4] https://www.nature.com/articles/s41598-023-40364-6

This PhD project seeks to revolutionize medical imaging by combining the power of citizen science with artificial intelligence (AI). By involving citizens in the annotation of medical datasets, this approach breaks away from traditional limitations, speeding up the annotation process and enriching the data with diverse perspectives. This project not only drives innovation in AI but also empowers individuals to contribute to healthcare advancements, creating a valuable dataset that supports more adaptable AI models. Beyond the technical outcomes, it fosters a more informed and engaged public in the realm of healthcare innovation. In essence, this PhD represents a transformative fusion of citizen science and AI, addressing pressing societal challenges and driving collective progress in healthcare.

The project is structured into three main phases. First, it will engage citizens from diverse backgrounds to identify patterns that may be overlooked by clinical experts alone. Through this collaboration, a large, annotated dataset will be created, addressing the common challenge of limited labeled medical images and enabling more robust AI development. The second phase focuses on algorithm generalization, where insights from citizen annotations help develop more inclusive and generalized AI models that work effectively across varied demographics and medical conditions. Finally, the project will promote public participation in healthcare innovation, raising awareness about AI in healthcare and ensuring ethical standards are met throughout.

This project focuses on studying possible links between psychology and artificial intelligence (AI), with a particular focus on large language models (LLMs). We will explore ways in which these research areas can connect and benefit from each other. 

This thesis proposal targets a significant practical challenge: efficiently governing the vast majority of AI systems that fall outside the "high-risk" category under the EU AI Act. While high-risk systems have clear obligations, the governance of minimal-risk and limited-risk systems (such as chatbots or automatic recommender systems) often lacks standardized, scalable methods.
The thesis proposal focuses on developing a Lightweight, Scalable Governance Framework and a Pragmatic Toolset for organizations to demonstrate compliance and responsible deployment for low-risk AI.
It will address questions such as:

• 1) What are the minimum necessary documentation, transparency, and human oversight requirements for limited-risk AI systems (e.g., recommender systems, chatbots) to meet the spirit of the EU AI Act without over-engineering compliance?
• 2) How can self-assessment methodologies be standardized and validated for minimal-risk AI to ensure organizational accountability?
• 3)Can a 'Risk-Migration' model be designed to dynamically reassess and transition low-risk systems to higher-risk tiers if their use context changes or unintended harms are detected post-deployment?

It concerns the following expertise areas:

• AI Risk & Compliance, Responsible AI Framework Design, Strategic Policy Advocacy.

It will contribute to a validated compliance playbook that reduces friction for innovation in low-risk AI while maintaining a baseline of ethical and legal integrity.