Yuyang Liu

FeCAM: Exploiting the Heterogeneity of Class Distributions in Exemplar-Free Continual Learning

2023 · arXiv (Cornell University)

Exemplar-free class-incremental learning (CIL) poses several challenges since it prohibits the rehearsal of data from previous tasks and thus suffers from catastrophic forgetting. Recent approaches to incrementally learning the classifier by freezing the feature extractor …

AsFT: Anchoring Safety During LLM Fine-Tuning Within Narrow Safety Basin

2025 · arXiv (Cornell University)

Fine-tuning large language models (LLMs) improves performance but introduces critical safety vulnerabilities: even minimal harmful data can severely compromise safety measures. We observe that perturbations orthogonal to the alignment direction - defined by weight differences …