1. events
  2. all
  3. 2024

Dec 9 (Mon) @ 2:00pm: "Towards Responsible ML: Enhancing Robustness of Deep Neural Networks and Promoting Fairness in Decision-Making," Bhagyashree Puranik, ECE PhD Defense

Date and Time

Location: Engineering Science Bldg (ESB), Room 2001
Zoom Meeting: https://ucsb.zoom.us/j/81824122259
Research Areas: Communications & Signal Processing
Research Keywords: Machine Learning, Responsible AI

Abstract

Designing trustworthy machine learning systems requires addressing critical challenges in robustness and fairness. Robust ML models must withstand corruptions, adversarial perturbations and other uncertainties to ensure reliable performance, while fairness mechanisms aim to mitigate biases and promote equitable outcomes across diverse groups in the population. This talk will focus on a novel approach for enhancing the robustness of deep neural networks, achieved through neuronal competition and a new "tilted exponential" objective function. The approach offers a significant step forward in robustness without relying on extensive data augmentation, demonstrating improved robustness across various image datasets. The talk will also briefly describe a complementary work addressing long-term fairness in sequential decision-making, which examines the hypothesis that slightly favoring underrepresented groups in selection can generate positive reinforcement.

Bio

Bhagyashree Puranik is a PhD candidate in the Department of Electrical and Computer Engineering at UC Santa Barbara, under the supervision of Prof. Upamanyu Madhow and Prof. Ramtin Pedarsani. She received her B.E. in Electronics and Communication Engineering from PES Institute of Technology, Bengaluru, in 2015, and her M.E. in Electrical Communication Engineering from the Indian Institute of Science, Bengaluru, in 2017. She is a recipient of the Regents Fellowship at UC Santa Barbara. Her research interests include designing robust machine learning models, fairness in decision-making, and representation learning.

Hosted By: Prof. Upamanyu Madhow

Submitted By: Bhagyashree Puranik <bpuranik@ucsb.edu>