Can Memory-Augmented Language Models Generalize on Reasoning-in-a-Haystack Tasks?
This addresses the problem of improving context processing for reasoning in language models, though it is incremental as it builds on existing memory-augmented approaches.
The paper tackles the brittleness of large language models in multi-hop reasoning tasks by proposing MemReasoner, a memory-augmented architecture that learns fact order and enables hopping, achieving strong generalization on synthetic tasks with minimal supervision (none for one-hop and 1% for two-hop tasks).
Large language models often expose their brittleness in reasoning tasks, especially while executing long chains of reasoning over context. We propose MemReasoner, a new and simple memory-augmented LLM architecture, in which the memory learns the relative order of facts in context, and enables hopping over them, while the decoder selectively attends to the memory. MemReasoner is trained end-to-end, with optional supporting fact supervision of varying degrees. We train MemReasoner, along with existing memory-augmented transformer models and a state-space model, on two distinct synthetic multi-hop reasoning tasks. Experiments performed under a variety of challenging scenarios, including the presence of long distractor text or target answer changes in test set, show strong generalization of MemReasoner on both single- and two-hop tasks. This generalization of MemReasoner is achieved using none-to-weak supporting fact supervision (using none and 1\% of supporting facts for one- and two-hop tasks, respectively). In contrast, baseline models overall struggle to generalize and benefit far less from using full supporting fact supervision. The results highlight the importance of explicit memory mechanisms, combined with additional weak supervision, for improving large language model's context processing ability toward reasoning tasks.