#catastrophic forgetting

A PyTorch tutorial demonstrating how to build a memory-augmented agent that uses differentiable memory, prioritized experience replay and meta-learning to adapt continuously without catastrophic forgetting.

'MIT shows that on-policy reinforcement learning preserves prior capabilities better than supervised fine-tuning by minimizing forward KL divergence between the base and fine-tuned models.'

EPFL researchers have developed MEMOIR, a novel framework that enables continuous, reliable, and localized updates in large language models, outperforming existing methods in various benchmarks.