Look Back to Reason Forward: Revisitable Memory for Long-Context LLM Agents

Researchers introduce ReMemR1, a new approach to improve large language models' ability to handle long-context question answering by integrating memory retrieval into the memory update process. The system enables non-linear reasoning through selective callback of historical memories and uses multi-level reward design to strengthen training.

• →ReMemR1 addresses key challenges in long-context LLM processing including evidence pruning, information loss, and sparse learning signals.
• →The system integrates memory retrieval into memory updates, allowing agents to selectively access historical memories for complex reasoning.
• →Multi-level reward design combines final-answer rewards with step-level signals to guide effective memory usage during training.
• →Experimental results show ReMemR1 significantly outperforms existing baselines while maintaining negligible computational overhead.
• →The approach specifically targets multi-hop reasoning tasks where evidence may be dispersed across millions of tokens.