MoRAM Solves LLM Catastrophic Forgetting with Rank-1 Memory Atoms

量子位 · wechat · 2026-08-04

A team from UNSW proposed MoRAM, a new architecture that decomposes model updates into minimal rank-1 "memory atoms" to effectively solve catastrophic forgetting in continual learning. The paper has been accepted by ICML 2026.

Original post →

More from Research

Research channel →