28th of May at 16h00, Gonçalo Carvalho and Jiawei Wang will give two short presentations, to promote discussion on two relevant ongoing or disruptive topics. Afterwards, there will be a social gathering where everyone can talk freely on whatever subjects they like.
Location: G4.1
Jiawei Wang – “Mitigating dataset biases for AI-based safety-critical perception systems”
Bio
Jiawei Wang is a Ph.D. student at CISUC, University of Coimbra. She received her master’s degree in Software Engineering from Beijing Institute of Technology, China, in 2020, with a specialization in Machine Learning applied to perception tasks. Under the supervision of Prof. João Campos, her current research is centered on characterizing and enhancing safety of AI-based perception systems by improving model generalization under dynamic working environments.
Abstract
AI is increasingly used in safety-critical perception tasks due to its exceptional performance on complex vision tasks. However, it reveals limited robustness under distributional shifts where the dataset bias is a key factor. In this talk, I will introduce our recent work on identifying and mitigating image dataset biases with the pedestrian classification as a case study. Through analysis of 4 large-scale datasets, we found the persistence of dataset biases and localized them in bias-rich image regions using explainable AI techniques (e.g., GradCAM and LayerCAM). Our ongoing work will focus on mitigating dataset biases with Image-to-Image techniques, targeting reducing biases without impairing useful information. The ultimate goal is to enhance the safety of AI perception systems by improving generalization across diverse real-world scenarios.
Gonçalo Carvalho – “Assisted generation and deployment of distributed data management systems”
Bio
Gonçalo Carvalho started his Ph.D. in 2018 in Data Models for multi-layer distributed databases.Since 2023, he has been working as a data engineering consultant.His research interests include databases, data modeling, distributed systems, cloud and edge computing.
