Continual Learning Foundation Models

Lifelong adaptation, rehearsal-free learning, and unlearning for evolving foundation models.

Goals

To enable foundation models to learn continuously from new data while retaining prior knowledge and forgetting sensitive information when required.

Overview

This direction studies how large models adapt over time without catastrophic forgetting. Research covers rehearsal-free continual learning, document retrieval under distribution shift, and principled machine unlearning for safe AI.

Key objectives

Design rehearsal-free continual learning for large models
Prevent catastrophic forgetting during sequential adaptation
Enable reliable machine unlearning in foundation models
Support lifelong document and knowledge retrieval

Key topics

Continual and lifelong learning
Rehearsal-free adaptation
Machine unlearning
Document retrieval under distribution shift

Papers in this direction

No papers listed yet.