STUDENTS’ PROJECTS

Our laboratory is offering an internship opportunity for students interested in artificial intelligence-based speech recognition and generation technologies. The project is part of an interdisciplinary research context involving speech recognition, artificial intelligence and psychology, with particular reference to the study of the correlations between vocal characteristics and mental disorders.

The objective of the internship is to create an audio dataset consisting of artificially generated speeches and stories, to be used in research on the recognition of mental pathologies through speech analysis. The activity will be divided into several phases: an analysis of the state of the art on speech generation and recognition models, the experimentation of AI-based tools for the generation of textual and vocal content, the design of effective prompts and the production of the final dataset.

The internship is mainly aimed at students of the master's degree in Communication Theory and Technology (TTC) with an interest in speech recognition, artificial intelligence and applications in the field of psychology. No specific previous training is required, but curiosity, ability to experiment and interest in new technologies are appreciated. It is possible to integrate the internship with a master's thesis project.

For more information, please contact Claudia Rabaioli.

Our laboratory is offering an internship opportunity for students interested in exploring the role of artificial intelligence in the generation of artistic content aimed at stimulating emotions. The project is part of an interdisciplinary context that combines art, artificial intelligence and psychology, with the aim of developing a set of mainly visual content capable of eliciting specific emotional responses.

The internship includes an initial phase of analyzing the state of the art on AI-based artistic generation models and existing techniques, followed by experimentation with generative tools, with a focus on designing effective prompts. The final phase will consist of creating and evaluating the generated multimedia content.

The activity is mainly aimed at students of the master's degree in Communication Theory and Technology (TTC), with an interest in generative art, artificial intelligence and the psychology of emotions. No previous training in specific fields is required, but curiosity, the ability to experiment and an interest in new technologies are appreciated. It is possible to integrate the internship with a master's thesis project.

For more information, please contact Claudia Rabaioli.

Music is usually employed as a stimulus to induce emotions in a listener.

Induced emotions are strongly dependent from the listener's musical preferences, their mood while listening to a song, their experience, and numerous other factors.

The appreciation for a specific song could be bounded non only to subjective factors, but also to objective characteristics of the song.
What is the general variation of the song? Is it instrumental or is there a singer giving voice to the melody? Are the lyrics understandable?

The proposed activities have two objectives: (i) prepare a questionnaire to allow volunteers to evaluate subjectively the complexity of popular songs, while considering the variation of their complexity over time, and (ii) devise some objective measures to evaluate if there is a link between the complexity of the audio signal (instrumental) and the listener preferences.erazione delle variazioni di complessità nel tempo.

The student will work on a dataset presenting the top 10 songs for each year starting from the Fifties.

The student will have the opportunity to work with a multidisciplinary group constituted by experts in the analyses of multimedia signals, Prof. Francesca Gasparini and Dr. Aurora Saibene of the MMSP lab, and experts in the field of psychology and neuroscience of music, Dr. Laura Ferreri and Dr. Carlotta Lega of the University of Pavia.

For more information, please contact Aurora Saibene.

The goal of this project is to analyze a public dataset of electronic health records of 425 patients diagnosed with depression who also suffered from heart failure. By applying computational statistics and supervised machine learning methods for feature ranking, we would like to detect what clinical features result being more correlated with heart failure, and which clinical features result being more correlated with depression. Computational methods will include traditional biostatistics tests (Student's t-test, Mann-Whitney U test, chi-squared test, etc.), and supervised machine learning approaches (for example, Random Forests, XGBoost, or others) for feature ranking. The feature ranking step will be carried out through feature recursive elimination (RFE), Shapley feature importance method, or others. The project will be carried out through open source programming languages such as R or Python. In case of interesting results, we will consider the possibility to write a scientific article to submit to journals such as BMC Medical Informatics or BioData Mining, or others.

For more information contact Davide Chicco.

Electroencephalography (EEG) has been exploited in rehabilitation systems of lower and upper limbs, having that the response from the motor cortex area presents specific synchronization and desynchronization in specific frequency ranges, and that the rehearsal of movements present the same neural patterns when compared with their executed counterparts.

Galvanic Skin Response (GSR) and photoplethysmographic (PPG) signals are good indicators of cognitive overload and stress, which can arise while performing motor imagery (MI) experiments due to the difficulty of these tasks and the necessity of repeating these activities multiple times to develop reliable Brain-Computer Interfaces (BCIs).

This thesis proposal has three main objectives: 1) find a literature corpus regarding MI in the EEG domain and understand how the use of GSR and PPG signals may improve MI-based BCIs, 2) design an experimental protocol for a preliminary study on single upper limb movement (e.g., hand grasping and flexing), exploiting EEG, GSR and PPG wearable devices, and 3) propose a second experimental session during which a feedback will be provided to the subjects while imaging the required movements in form of a simple game.

For more information contact Aurora Saibene.

While emotions are neural responses to specific feeling or external stimuli, the mood is a persisting state that is not necessarily tide to a specific stimulus and that may present different feelings and evoked emotions. However, a clear and well-established definition of mood in respect to emotions and feelings has yet to be found.

Different physiological signals have been demonstrated to be associated with specific emotions, while virtual reality (VR) environments have proved their ability to elicit specific emotions.

This thesis proposal has three main objectives: 1) moving from emotion to mood detection considering the use of wearable sensing technologies, 2) define specific scenarios to reproduce in VR environments, 3) provide a tentative analysis of the data derived from experimental sessions with the main objective of providing a mood manipulation.

This thesis will be performed in the context of an open collaboration with Professor Francesco Ferrise of the Department of Mechanical Engineering, Politecnico di Milano.

For more information contact Francesca Gasparini and Aurora Saibene.

Electroencephalography (EEG) has been exploited in rehabilitation systems of lower and upper limbs, having that the response from the motor cortex area presents specific synchronization and desynchronization in specific frequency ranges. Unfortunately, EEG signals can record only the surface activity and may be unable to provide evidence regarding intricate movements.

Electromyographic (EMG) signals are instead able to acquire the muscular activity with more precision and may thus represent an ally to provide a more in-depth assessment of movements.

This thesis proposal has three main objectives: 1) find a literature corpus regarding EEG and EMG signals (used separately or considering sensor fusion) used for upper or lower limb movement recognition, 2) design an experimental protocol for a preliminary study on single upper or lower limb movement (e.g., foot flexion, hand gestures), exploiting EEG and EMG wearable devices, and 3) propose a second experimental session during which a feedback will be provided to the subjects while executing or imaging the required movements.

For more information contact Aurora Saibene.

Emotion recognition has been drawing attention from multidisciplinary research communities and has been relying increasingly on artificial intelligence strategies to better model dependencies between cognition and physiological processes.

Physiological signals like Galvanic Skin Response (GSR) have revealed to be precise indicators of emotions, but also electroencephalographic (EEG) and photoplethysmographic (PPG) signals have been exploited in emotion-related experimentations.

This thesis proposal has three main objectives: 1) exploit multi-modal datasets from the literature (e.g., DEAP) to propose new emotion-recognition models, 2) record GSR, EEG, and PPG signals using wearable devices while subjects undergo an emotional engaging experiment, and 3) provide a preliminary assessment of the acquired data on a population subset.

For more information contact Alessandra Grossi and Aurora Saibene.

Stress detection has been drawing attention from multidisciplinary research communities, especially concerning mental health care, and has been relying increasingly on artificial intelligence strategies to better model dependencies between cognition and physiological processes.

Physiological signals like Galvanic Skin Response (GSR) and data from the photoplethysmogram (PPG) have revealed to be precise indicators of stress, but also electroencephalographic (EEG) signals have been exploited in cognitive overload experimentations.

This thesis proposal has three main objectives: 1) explore the literature for datasets using one or all the presented modalities, 2) record GSR, EEG, and PPG signals using wearable devices while subjects undergo a stress and relax inducing experiment, and 3) provide a preliminary assessment of the acquired data on a population subset.

For more information contact Aurora Saibene.

The project purpose is to analyse a subject physiological signal in a stressing environment, thus the experiment takes place on a two-way transit street (not controlled by traffic lights).

The experimental protocol its repeated for 4 trials and is described in the following:

• 1 minute baseline recording;
• the volunteer walks safely back and forth om the side walk for 30 + 30 meters;
• a resting time is given to the volunteer;
• 1 minute baseline recording;
• the volunteer crosses the street two times to return to the starting point;
• a resting time is given to the volunteer.

The project purpose is to analyse if there is a difference in a subject physiological response while he/she is listening to a heterogeneous audio playlist. This playlist is composed by tracks labelled as stressful or relaxing.

The experimental protocol is described in the following:

• 7 stressful and 7 relaxing audio tracks are randomly selected from a playlist;
• 150 seconds baseline recording;
• the volunteer listen to a 60 seconds track;
• 15 seconds are given to the volunteer to judge the track as stressful or relaxing.

The last two points are repeated for all the selected tracks.

The data recorded through Shimmer3 GSR+ (http://www.shimmersensing.com/) are: heart rate and GSR.

The project purpose is to analyse the subjective response of volunteers exposed to a heterogeneous experimental paradigm through means of electroencephalography.
The data are collected through the Unicorn Hybrid Black (g.tec medical engineering GmbH) using it with dry electrodes to assess the possibility of having more comfortable devices for electroencephalographic signal recording.