Campus Event Calendar
Event Entry
What and Who
Robustness Against Spurious Features
Hosna Oyarhoseini
Sharif University of Technology
PhD Application Talk
AG 1, AG 2, AG 3, INET, AG 4, AG 5, D6, SWS, RG1, MMCI
AG Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Tuesday, 4 February 2025
11:30
30 Minutes
Virtual talk
zoom
Abstract
AI is rapidly advancing and transforming industries, but its full potential in real-world applications is often hindered by issues of reliability and trustworthiness. A critical challenge lies in addressing spurious correlations—patterns in data that models rely on for predictions but are not truly relevant to the task. These spurious features can lead to significant failures, especially when models encounter distributional shifts.
In this talk, I will discuss my ongoing research, which focuses on demonstrating that widely used models, particularly vision-language models like CLIP, often lack robustness against spurious correlation even in real-world applications. I also explore how to enhance the robustness and generalization of AI systems by efficiently mitigating reliance on spurious features. Additionally, I will share what inspired me to pursue this concern and outline broader research directions I hope to explore in the future to contribute to the development of dependable AI systems.
In this talk, I will discuss my ongoing research, which focuses on demonstrating that widely used models, particularly vision-language models like CLIP, often lack robustness against spurious correlation even in real-world applications. I also explore how to enhance the robustness and generalization of AI systems by efficiently mitigating reliance on spurious features. Additionally, I will share what inspired me to pursue this concern and outline broader research directions I hope to explore in the future to contribute to the development of dependable AI systems.
Contact
Ina Geisler
+49 681 9325 1802
--email hidden
Virtual Meeting Details
Zoom
passcode not visible
logged in users only
Ina Geisler, 01/24/2025 14:42 -- Created document.