Ofertas de proyectos

Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, a quantum computer faces many challenges relative to the movement of qubits which is completely different from the movement of classical data. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems and the algorithms that run within them, following the current European projects at N3Cat (www.n3cat.upc.edu) on scalable quantum computing.

The interested candidate will work in a group of several PhD students and in collaboration with Universitat Politècnica de València, working in ONE of the following areas:

• Study the use of AI to model quantum computers or to assist in the development of architectural methods (qubit mapping, qubit partitioning, qubit routing).
• Study the potential scaling and adaptation of the existing quantum computing architectures to specific quantum machine learning algorithms (QAOA, VQE, others).

Proposal 1: Automated Detection of Cognitive Distortions: Enhancing Mental Health Diagnosis in Spanish-Speaking Populations

Background

Depression and anxiety are among the most significant health issues worldwide, affecting up to 50% of the population during their lifetime (Santomauro et al., 2021). Despite the effectiveness of psychological interventions in reducing their impact, these conditions are often underdiagnosed and undertreated, primarily due to limited human resources. When individuals are depressed or anxious, their information processing tends to be biased, leading to "cognitive distortions" For example, a common thought among depressed individuals is: "I am a failure." The development of Natural Language Processing (NLP) offers a promising opportunity to improve the detection of these cognitive distortions automatically. Previous studies have aimed at identifying cognitive distortions using posts from online mental health platforms, annotated by mental health professionals (Simms et al., 2017; Rojas-Barahona et al., 2018; Shreevastava and Foltz, 2021) and people lacking clinical experience (Shickel et al., 2019). However, these datasets do not represent individuals with depressive or anxiety disorders in clinical settings. Furthermore, in the Spanish language, there is a significant lack of data available for such analysis.

Aim

To train automated algorithms to identify "cognitive distortions" from clinical data in the Spanish language.

Methods

1. Generation of the Dataset: Expert clinical psychologists from the Hospital Clínic de Barcelona have generated 1500 quotes containing ten types of "cognitive distortions" inspired by patients with depression and anxiety (Table 1). As control quotes, the psychologists have also produced non-distorted versions of the same thoughts, known as "alternative thoughts" (Beck, 2005).

1. Annotation of the Dataset: Clinical psychologists, who were blind to the generated dataset, were presented with random quotes and instructed to determine whether each quote contained cognitive distortions using binary distorted/non-distorted labels. If a quote was considered "distorted," they were then asked to identify the type of cognitive distortion present from among the ten predefined types. Given that cognitive distortions can overlap (i.e., a single quote can exhibit multiple types of distortions), annotators were allowed to select multiple types of distortion. To ensure a high-quality dataset, a representative sub-corpus (10%) of the dataset was first annotated. Inter-annotator agreement metrics were calculated, and the annotation system was updated based on suggestions from the annotators. Once the initial 10% was satisfactorily annotated, the remaining 90% of the dataset was annotated. Disagreements during this phase were debated among the clinical psychologists, and a consensus was reached whenever possible to finalize the annotations.

Plan

1. Detect Cognitive distortions: Develop algorithms to distinguish between cognitive distortions and non-distorted thinking. High specificity is crucial to avoid invalidating patients' emotions, which could be counterproductive for psychotherapeutic intervention.

1. Detect the Type of Cognitive distortions: The first step in any psychotherapeutic intervention is psychoeducation, which involves informing patients about their symptoms and diagnoses. For cognitive therapy, this includes identifying and explaining the types of cognitive distortions. Therefore, it is important to accurately identify both the presence and type of cognitive distortions.

Various NLP and machine learning methods will be tested for both purposes. Classification accuracies will be compared, and the best-performing algorithm will be selected for each task. The specific language cues used for classification will be studied to understand and make the classification systems transparent for clinical applicability.

Our previous work on the same dataset, involving uncovering structural and emotional patterns in cognitive distortions using NLP and cognitive network science (Molins et al., in preparation), revealed several key findings. We found that cognitive distortions, compared to alternative thoughts, contained more words, exhibited increased negative valence and excitatory arousal, showed higher structural imbalance and less compartmentalization, and were characterized by higher incoherence. Cognitive distortions were also associated with more negative emotions, such as anger, disgust, and sadness. Additionally, different types of cognitive distortions displayed distinct characteristics: for example, catastrophism was associated with higher structural imbalance, different central actors, and increased sadness, while labeling was linked to higher levels of disgust. This nuanced understanding can be leveraged to train more effective detection algorithms.

Impact

Developing automated models for detecting cognitive distortions will likely enhance the identification of individuals experiencing depression and anxiety. These models can be applied to social media platforms (Ramírez-Cifuentes et al., 2020) and mental health chatbots (Anmella et al., 2023), which are increasingly used worldwide. By leveraging these technologies, psychotherapeutic interventions can be provided to a broader population, addressing the current shortage of psychologists and mental health professionals. This development is particularly crucial for the Spanish-speaking community, where such models are currently lacking.

References

Anmella, G. et al. (2023) 'Vickybot, a Chatbot for Anxiety-Depressive Symptoms and Work-Related Burnout in Primary Care and Health Care Professionals: Development, Feasibility, and Potential Effectiveness Studies', J Med Internet Res 2023;25:e43293 https://www.jmir.org/2023/1/e43293, 25(1), p. e43293. Available at: https://doi.org/10.2196/43293.

Beck, A.T. (2005) 'The current state of cognitive therapy: a 40-year retrospective', Archives of general psychiatry, 62(9), pp. 953¿959. Available at: https://doi.org/10.1001/ARCHPSYC.62.9.953.

Ramírez-Cifuentes, D. et al. (2020) 'Detection of Suicidal Ideation on Social Media: Multimodal, Relational, and Behavioral Analysis', J Med Internet Res 2020;22(7):e17758 https://www.jmir.org/2020/7/e17758, 22(7), p. e17758. Available at: https://doi.org/10.2196/17758.

Rojas-Barahona, L. et al. (2018) 'Deep learning for language understanding of mental health concepts derived from Cognitive Behavioural Therapy', pp. 44¿54. Available at: https://doi.org/10.18653/v1/w18-5606.

Santomauro, D.F. et al. (2021) 'Global prevalence and burden of depressive and anxiety disorders in 204 countries and territories in 2020 due to the COVID-19 pandemic', Lancet (London, England), 398(10312), pp. 1700¿1712. Available at: https://doi.org/10.1016/S0140-6736(21)02143-7/ATTACHMENT/927FDFEF-CCD4-4655-AACF-4E7D54DFECF5/MMC1.PDF.

Shickel, B. et al. (2019) 'Automatic Detection and Classification of Cognitive Distortions in Mental Health Text', Proceedings - IEEE 20th International Conference on Bioinformatics and Bioengineering, BIBE 2020, pp. 275¿280. Available at: https://doi.org/10.1109/BIBE50027.2020.00052.

Shreevastava, S. and Foltz, P.W. (2021) 'Detecting Cognitive Distortions from Patient-Therapist Interactions', Computational Linguistics and Clinical Psychology: Improving Access, CLPsych 2021 - Proceedings of the 7th Workshop, in conjunction with NAACL 2021, pp. 151¿158. Available at: https://doi.org/10.18653/V1/2021.CLPSYCH-1.17.

Simms, T. et al. (2017) 'Detecting Cognitive Distortions Through Machine Learning Text Analytics', Proceedings - 2017 IEEE International Conference on Healthcare Informatics, ICHI 2017, pp. 508¿512. Available at: https://doi.org/10.1109/ICHI.2017.39.

What is Emergence? Emergence refers to the phenomenon where a complex system exhibits properties or behaviors that its individual components do not possess in isolation. These emergent features arise only when the components interact within a broader context. In philosophy, science, and art, emergence plays a pivotal role in theories related to integrative levels and complex systems. For example, life, as studied in biology, emerges from the underlying chemistry and physics of biological processes.

Emergence in Large Language Models Recent research has highlighted emergent behavior in Large Language Models (LLMs). These models, such as GPT-3, exhibit surprising capabilities beyond their individual components (words or tokens). The interactions between countless parameters give rise to emergent linguistic abilities, including natural language understanding, generation, and context-based reasoning. For further details, refer to https://arxiv.org/pdf/2206.07682

What is Assembly Theory (AT)? Assembly Theory (AT) provides a novel framework for quantifying complexity without altering fundamental physical laws. Unlike traditional point-particle models, AT defines objects based on their potential formation histories. These "objects" can exhibit evidence of selection within well-defined boundaries. AT allows us to explore emergent properties by considering how components assemble into coherent entities, shedding light on the intricate dynamics of complex systems. See this paper for further information: https://www.nature.com/articles/s41586-023-06600-9

Research Goals: This thesis aims to apply Assembly Theory to understand emergence in Large Language Models. We will test AT in this particular set emergence problem.

Proposal 2: Automated Speech Analysis in Bipolar Disorder: Enhancing Diagnosis and Monitoring.

Background

Bipolar Disorder (BD) is a psychiatric condition in which people experience significant shifts in mood, energy, and thought processes during manic and depressive episodes (Nierenberg et al., 2023). Manic episodes involve heightened energy, rapid speech, and grandiose thoughts, while depressive episodes are marked by low energy, slow speech, and feelings of hopelessness.

Language, expressed through speech, provides a privileged window into the mind and is thus a cornerstone of psychiatric evaluation. During these evaluations, clinicians routinely assess speech features such as (A) acoustic properties, (B) formal aspects, (C) language content, and (D) emotionality, albeit subjectively.

1. Acoustic features (Table 1) (e.g., pitch and volume) reflect mood states (Low, Bentley and Ghosh, 2020). Manic episodes might be characterized by increased pitch and loudness, while depressive episodes may present with reduced pitch variation and monotonous speech.
2. Formal aspects (Table 2) (e.g., flow, fluency, rhythm, quantity, and latency) can delineate distinct patterns associated with different phases (euthymia, mania, depression) of BD (Weiner et al., 2019). Overly rapid speech may indicate manic thought patterns, whereas slowed, halting speech can suggest depressive cognition.
3. Language content features (Table 3) (e.g., coherence and cognition) reflects thought organization (Iter, Yoon and Jurafsky, 2018). Manic episodes may involve tangential or circumstantial speech, while depressive episodes often include pervasive negative thoughts.
4. Emotionality features (Table 4) (e.g., emotional valence and content) helps differentiate manic and depressive phases (Khorram et al., 2018). Manic speech often conveys exaggerated positive emotions, whereas depressive speech typically reflects sadness and despair.

Modern technology enables high-fidelity speech recording and automated analysis of these features (DeSouza et al., 2021), showing potential in diagnosing and monitoring BD (Guidi et al., 2015; Faurholt-Jepsen et al., 2016).

We hypothesized that (i) speech features will correlate with the severity of manic and depressive symptoms, (ii) they will effectively differentiate between manic, depressive, and euthymic phases in BD, as well as between mania/depression and response/remission, (iii) only specific speech features and speech tasks will be relevant for each of these analyses.

Aims

1. To correlate speech features with manic-depressive symptoms severity in BDf
2. To use speech features to develop predictive models for diagnostic (i.e., manic, depressive, and euthymic phases in BD), and treatment outcomes (i.e., acute phases of mania/depression vs response/remission)
3. To determine which specific speech features and speech tasks are most relevant for each of the previous analyses.

Methods

A naturalistic, observational study was conducted. Patients with BD experiencing manic and depressive episodes underwent longitudinal audio recording during acute phases and after response/remission using a dual-microphone setup. Patients during euthymia (mood stability) were recorded once. Interviews included clinical evaluation, cognitive tasks, standard text reading, and emotional and non-emotional storytelling (Figure 1).

Audio recordings from 76 patients (24 manic, 21 depressed, 31 euthymic) were collected. Among the patients in the acute phase, 17 manic patients and 9 depressed patients were recorded longitudinally at clinical remission. Sociodemographic and clinical data of the sample are reported in Table 5. The average age was 48.1±13.3 years, and 64.6% were female. The mean Young Mania Rating Scale (YMRS) score for manic patients was 24±8.5 (moderate-to-severe) during acute episodes, reducing to 5.9±6.2 (mild-to-minimal symptoms) after response/remission. Depressed patients had a mean Hamilton Depression Rating Scale (HDRS-17) score of 17.1±4.4 (moderate-to-severe) during acute episodes, decreasing to 3.3±2.8 (mild-to-minimal symptoms) after remission. Euthymic patients exhibited mild-to-minimal symptoms (mean YMRS score of 0.97±1.4 and an HDRS-17 score of 3.9±2.9).

Recordings were automatically diarized and transcribed.

Plan

1. To extract speech features from the audio recordings, including (A) acoustic properties, (B) language content, (C) formal aspects, and (D) emotionality.
2. Statistical analyses:

1. To correlate speech features with clinical scales, 
2. To use lasso logistic regression to develop predictive models for diagnostic and treatment outcomes using speech features, and 
3. To use variable relevance methods to identify the most pertinent speech features for these analyses, and assessing the magnitude of correlation and prediction accuracies across different speech tasks.

Impact: Automated speech analysis in BD might provide objective, quantitative markers for psychopathological (manic/depressive) alterations. This technology could potentially identify subtle alterations imperceptible to clinicians, signaling early signs of acute relapse and allowing for early intervention. Implementing this technology could improve diagnosis, monitoring, and prediction of treatment response.

References

DeSouza, D.D. et al. (2021) 'Natural Language Processing as an Emerging Tool to Detect Late-Life Depression', Frontiers in Psychiatry. Frontiers Media S.A. Available at: https://doi.org/10.3389/fpsyt.2021.719125.

Faurholt-Jepsen, M. et al. (2016) 'Voice analysis as an objective state marker in bipolar disorder', Translational psychiatry, 6, p. e856. Available at: https://doi.org/10.1038/tp.2016.123.

Guidi, A. et al. (2015) 'Voice quality in patients suffering from bipolar disease', in Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS. Institute of Electrical and Electronics Engineers Inc., pp. 6106¿6109. Available at: https://doi.org/10.1109/EMBC.2015.7319785.

Iter, D., Yoon, J. and Jurafsky, D. (2018) 'Automatic Detection of Incoherent Speech for Diagnosing Schizophrenia', in. Association for Computational Linguistics (ACL), pp. 136¿146. Available at: https://doi.org/10.18653/v1/w18-0615.

Khorram, S. et al. (2018) 'The PRIORI Emotion Dataset: Linking Mood to Emotion Detected In-the-Wild'. Available at: http://arxiv.org/abs/1806.10658.

Low, D.M., Bentley, K.H. and Ghosh, S.S. (2020) 'Automated assessment of psychiatric disorders using speech: A systematic review', Laryngoscope Investigative Otolaryngology, 5(1), pp. 96¿116. Available at: https://doi.org/10.1002/lio2.354.

Nierenberg, A.A. et al. (2023) 'Diagnosis and Treatment of Bipolar Disorder: A Review', JAMA. American Medical Association, pp. 1370¿1380. Available at: https://doi.org/10.1001/jama.2023.18588.

Weiner, L. et al. (2019) 'Thought and language disturbance in bipolar disorder quantified via process-oriented verbal fluency measures', Scientific Reports, 9(1). Available at: https://doi.org/10.1038/s41598-019-50818-5.

Seeing that not all neural networks fit to all problems, and that relational data is present in a wide variety of aspects of our daily life, the main focus of this thesis in N3Cat (www.n3cat.upc.edu) and BNN-UPC (www.bnn.upc.edu) is to explore the possibilities of the Graph Neural Networks (GNNs), whose aim is to learn and model graph-structured relational data. We are looking for students willing to study the uses, architectures, and algorithms of GNNs. To this end, the candidate will work on ONE of the following areas:

• Uses: Applying GNNs in emerging application areas, including but not limited to (1) electroencephalogram (EEG) analysis for Alzheimer's disease detection, epilepsy classification, motor imagery; (2) acceleration of the compilation of quantum computing algorithms
• Architectures: Investigating ways to accelerate the processing of GNNs in multiple computing platforms (CPU, GPU, accelerators).
• Algorithms: Developing meta-learning data-driven models to estimate the accuracy of a GNN for a given graph, without training, through synthetic graph generation and correlation analysis.

Recent advancements in nanotechnology have enabled the development of means for sensing and wireless communications with unprecedented miniaturization and capabilities, to the point that they can be introduced into the gastrointestinal tract inside a pill or into the bloodstream in the form of passively flowing nanomachines. 

This opens the door to the idea of intra-body communication networks, this is, a swarm of nanosensors inside the human body that use communications to coordinate their actions to sense and localize specific events (lack of oxygen, biomarkers, etc). This can lead to the development of applications such as continuous monitoring of diabetes, detection and localization of cancer micro-tumors, or early detection of blood clots. These possibilities are currently investigated by our team at the N3Cat (www.n3cat.upc.edu).

In this context, we are looking for excellent and self-motivated individuals who are eager to work on developing AI schemes (based on graph neural networks or multi-agent RL) for the detection and localization of events inside of the human body. Data will be gathered with an in-house simulator that integrates mobility models (BloodVoyagerS) and communication models (TeraSim).

Large Language Models (LLMs) have rapidly emerged as a central component of modern AI technology, with hundreds of models released in recent years demonstrating unprecedented capabilities. While many LLMs initially focused on text-based tasks, recent advancements have expanded into multimodal approaches that integrate text with other data types, such as speech and images.

As the power and versatility of these models continue to grow, so do concerns about their security and ethical implications. A key issue in this area is the potential for LLMs to generate harmful or toxic content, even in response to seemingly benign inputs. Addressing this challenge requires robust benchmarks that can accurately evaluate model behavior under a wide range of conditions.

This project aims to reconstruct the MuTox benchmark, along with other emerging toxicity evaluation frameworks, and make them publicly available via an open-source repository hosted by UPC. In addition to the benchmark reconstruction, the project will involve an in-depth evaluation of the latest Large Language Models, such as LLaMA 3.1, using these newly developed benchmarks.

The project will provide an opportunity to collaborate with leading AI researchers from FAIR (Facebook AI Research) and Meta, enabling the student to work at the cutting edge of AI safety and security.

Traffic analysis is crucial for urban operations and planning, while the availability of dense urban traffic data beyond loop detectors is still scarce. A large-scale floating vehicle dataset of per-street segment traffic information, Metropolitan Segment Traffic Speeds from Massive Floating Car Data in 10 Cities (MeTS-10) is available for 10 global cities with a 15-minute resolution for collection periods ranging between 108 and 361 days in 2019¿2021 and covering more than 1500 square kilometers per metropolitan area. MeTS-10 enables novel, city-wide analysis of mobility and traffic patterns for ten major world cities, overcoming current limitations of spatially sparse vehicle detector data.  MeTS-10 features traffic speed information at all street levels from main arterials to local streets for Antwerp, Bangkok, Barcelona, Berlin, Chicago, Istanbul, London, Madrid, Melbourne, and Moscow. The dataset leverages the industrial-scale floating vehicle Traffic4cast data with speeds and vehicle counts provided in a privacy-preserving spatio-temporal aggregation. Data has to be map-matched to the OpenStreetMap (OSM) road graph. City datasets can be compared with other publicly available stationary vehicle detector data (for Berlin, London, and Madrid) and the Uber traffic speed dataset (for Barcelona, Berlin, and London). A comparison of the differences across some of the datasets in spatio-temporal coverage and variations in the reported traffic will be addressed in this project. The large spatial and temporal coverage offers an opportunity for joining the MeTS-10 with other datasets, such as traffic surveys in traffic planning studies or vehicle detector data in traffic control settings. A pipeline is already available to derive the dataset from the Traffic4cast data published by HERE. HERE is a technologal company providing a platform for the visualization and analysis of location data. To ensure data privacy, the Traffic4cast dataset was published as rasterized and aggregated cell-based data, nevertheless providing a high spatial and temporal resolution.

Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique that measures changes in oxygenated (HbO2) and deoxygenated hemoglobin (HbR) in the cerebral cortex. Due to its portability and low cost, fNIRS has been used in Brain-Computer Interface (BCI) applications, characterizing hemodynamic responses to varying stimuli, and investigating auditory and visual-spatial attention during Complex Scene Analysis (CSA). In this project, we want to design and implement an fNIRS study with a goal of studying the impact of neural and BCI outcomes to improving the training of LAI models' attention mechanism (e.g., Transformer attention) during reading comprehension tasks (e.g., the participants will be judging the quality of generated text). We want to demonstrate experimentally that augmenting a model with fNIRS data carries neural activity features complementing the information captured by the model and demonstrate that it improves the models' performance. To this end, we will have to collect data from participants and test how different Transformer models benefit from different types of fNIRS attention masks.

The candidate will:

• Carry out controlled studies and collect data from a number of participants.
• Try different approaches to incorporating fNIRS signal in the training process of DL/LLM models and compare against baselines/experiments (these should be replicated).
• Carry out ablation studies, personalisation techniques, error analyses, etc.
• Contribute to authoring a scientific article

Internet services are known to collect large amounts of personal information. As a result, more than half of EU citizens are concerned about their online privacy. In this context, data brokers are companies devoted to collecting and selling personal information to other companies. Data brokers are often implemented as third-party trackers, which allow them to gain visibility across the Internet. Recent works show this personal information is not only used for targeted advertising, but also for more obscure practices, such as price discrimination, credit scoring, phishing or identity theft. The project addresses these growing privacy concerns by enhancing ePrivo, a new online privacy observatory. ePrivo will continuously scan the Internet to unveil third-party trackers, the tracking methods they use and the data brokers behind them. A proof-of-concept prototype of ePrivo is already available at https://eprivo.eu. In this project, we will deploy the ePrivo service in production, extend its capabilities for web tracking and phishing detection, and release it under an open-source license. The results will be useful to the RIPE community, including ISPs, network operators, Internet users, policy makers and researchers.

The application of Artificial Intelligence (AI) and Machine Learning (ML) to network security (AI4SEC) is paramount against cybercrime. While AI/ML is mainstream in domains such as computer vision and natural language processing, traditional AI/ML has produced below-par results in AI4SEC. Solutions do not properly generalize, are ineffective in real deployments, and are vulnerable to adversarial attacks. A fundamental limitation is the lack of AI/ML technology specific to network security. Due to their unique ability to learn and generalize over graph-structured information, graph-learning approaches, and in particular Graph Neural Networks (GNNs), have recently enabled groundbreaking applications in multiple fields where data are generally represented as graphs. Network security data are intrinsically relational, and initial research suggests that graph-structured representations and GNNs have the potential to become foundational to AI4SEC, in the way convolutional and recursive networks were to computer vision and natural language processing.
The goal of GRAPHS4SEC is to leverage graph data representations and modern GNN technology to conceive a new breed of robust GNN-based network security methods which could radically advance the AI4SEC practice. The objectives of GRAPHS4SEC are: (a) to investigate algorithmic methods that facilitate modeling and learning from graph-based network security data; (b) to compare the benefits and overheads of GNN-based AI4SEC to traditional AI/ML in terms of detection performance, generalization, scalability, and robustness against adversarial attacks; (c) to showcase the benefits and improvements of GRAPHS4SEC technology in four critical, real-world network security applications with significant impact for society, considering (in particular) the detection and early mitigation of phishing and fake/malicious websites, a threat among the most popular and society-wide harmful in today's Internet.

The internship will be formalized as a CCE (Conveni de Cooperació Educativa).

To apply you need to send an email to pere.barlet@upc.edu with your CV and your bachelor and master transcripts.

Eye movement features are considered to be direct signals reflecting human attention distribution with a low cost to obtain, inspiring researchers to augment language models with eye-tracking (ET) data. In this project, we want to investigate how to operationalise eye tracking (ET) features, such as first fixation duration (FFD) and total reading time (TRT), as the cognitive signals to augment LAI models' attention mechanism (e.g., Transformer attention) during training. We want to demonstrate experimentally that augmenting a model with ET data carries linguistic features complementing the information captured by the model and demonstrate that it improves the models' performance. To this end, we will have to collect data from participants and test how different Transformer models benefit from different types of ET attention masks.

The candidate will:

• Carry out controlled studies and collect data from a number of participants.
• Try different approaches to incorporating gaze features in the training process of DL/LLM models and compare against baselines/experiments (these should be replicated).
• Carry out ablation studies, personalisation techniques, error analyses, etc.
• Contribute to authoring a scientific article