Doctoral Thesis: Robust Learning from Uncurated Data
32-G449 (Kiva)
Ching-Yao Chuang
Abstract:
The field of machine learning has witnessed growing interest in learning from uncurated data, which involves training models using data that has not been carefully curated or labeled. However, this type of data is typically noisy, incomplete, and riddled with errors, making it challenging for machine learning algorithms to learn effectively. My work focuses on the development of robust learning methods that can effectively leverage uncurated data while remaining resilient to the inherent noise and biases present in the data. Specifically, we investigate the robustness of contrastive learning, a prominent technique for self-supervised representation learning that contrasts semantically similar and dissimilar pairs of samples. We further extend our analysis to multi-modal learning, and develop an effective algorithm to remove inappropriate biases in foundational vision-language models.
Thesis Committee:
Stefanie Jegelka, Antonio Torralba, Phillip Isola
Details
- Date: Thursday, June 29
- Time: 2:00 pm - 3:30 pm
- Category: Thesis Defense
- Location: 32-G449 (Kiva)