Qinglin Meng

Qinglin Meng

PhD Candidate in Computer Science at Purdue University. Research in machine learning theory, reinforcement learning, and LLMs.

Representation Preserving Multiclass Agnostic to Realizable Reduction

Published in International Conference on Machine Learning (ICML 2025), 2025

We study multiclass agnostic to realizable reductions and characterize conditions under which such reductions can preserve the representation class. Our results provide theoretical foundations for understanding the relationship between agnostic and realizable learning in the multiclass setting.

Steve Hanneke, Qinglin Meng, Amirreza Shaeiri (alphabetical order)
Download Paper