Kini Chen (Neurology and Neurosurgery, Faculty of Medicine and Health Sciences)
Identifying Key Factors in Methodological Variability of Dynamic Functional Connectivity through Spectral Inference and Feature Selection
Dynamic functional connectivity (dFC), the variation of brain region functional connections over time measured with Brain Oxygen Level Dependent signals, has been a growing field of research due to its importance in understanding brain processes and potential applications as a biomarker for predicting or tracking the progression of neurological diseases. However, the choice of dFC assessment methodology has been found to significantly impact the dFC results, questioning their reliability. In this project, we aim to identify such factors by first finding functional connections for which dFC assessment is the most variable, and then characterize these connections using factors such as connection length, localisation in brain network and hemisphere association. We will then use the most robust connections to predict subjects鈥� brain state and characteristics such as disease status.
听
Hadrien Padilla (Computer Science, Faculty of Science)
Learning to Value: Hippocampal-Inspired Reward Encoding for Responsible AI System
In an era where AI systems are integral to societal decision-making, ensuring that these systems are ethically sound and equitable is paramount. It is important to integrate our values into AI systems to ensure ethical decision-making. Currently, values are embedded into these systems through engineering. It would be beneficial to build systems that are able to identify their own goals and values, as this would allow more understanding and control over shaping these goals and values. By integrating reward representations inspired by biological mechanisms of the hippocampus into predictive neural networks, my research aims to guide the responsible evolution of AI.
听
Nahian Rahman (Computer Science, Faculty of Arts)
Exploring Bias in AI-Driven Facial Impression Predictions: Ensuring Fairness in Automated Decision-Making
AI-driven facial analysis is increasingly used in hiring to assess traits like trustworthiness and competence, influencing real-world decisions. However, research shows that human facial impressions are shaped by group stereotypes. AI models trained on biased data may reinforce or amplify these biases. This poses risks of discriminatory hiring practices.听This project aims to evaluate how AI models predict facial impressions, and assess if these predictions reflect or amplify human biases. By comparing AI-generated facial impression predictions to human stereotypes, this project will identify correlations and disparities between the two. We aim to develop fairness metrics to benchmark AI models used in hiring. If time allows, we will also explore ways to reduce AI-driven bias in facial impression predictions.
听
Alina Shimizu-Jozi (Computer Science, Faculty of Science)
The Role of Gender in Medical AI: A study on Chatbot-Patient Interactions
This project aims to examine how user gender, perceived chatbot gender, and communication styles affect the quality and effectiveness of digital mental health tools. By randomizing users' assignment to a male- or female-presenting chatbot 鈥� or allowing them to choose 鈥� we will investigate the impact of gender concordance, user preferences, and whether choice itself influences outcomes.
听
Michael Yu (Computer Science, Faculty of Arts)
Bridging the Past and Future: Leveraging AI to Analyze Historical Weather Data for Climate Resilience
This research aims to leverage state-of-the-art large language models and historical weather data to enhance climate resilience strategies by uncovering societal vulnerabilities and adaptation methods from the past. The project integrates advanced Retrieval-Augmented Generation techniques with historical newspaper archives to extract meaningful insights about disruptive weather events, societal responses, and resilience strategies from the 19th century to the present.
