Search papers, labs, and topics across Lattice.
This study investigates the role of conditional memory in scientific reasoning by analyzing its effectiveness in retrieving specialized knowledge during multi-step computations. The authors introduce a Knowledge Boundary-Aware Router that dynamically determines when and how memory should be activated based on task-specific input proxies. Their experiments reveal that selective memory allocation significantly enhances reasoning performance while minimizing detrimental effects, underscoring the importance of context in memory utilization for language models.
Conditional memory can either enhance or hinder scientific reasoning, and knowing when to activate it is crucial for optimal performance.
Scientific reasoning requires language models to retrieve specialized knowledge and incorporate it reliably into multi-step computation. Conditional memory provides an explicit lookup pathway that complements dense neural representations, but its usefulness is inherently input- and computation-dependent: retrieved information may repair missing scientific associations, yet it may also introduce distracting shortcuts or interfere with reasoning that the base model can already perform correctly. In this work, we systematically investigate when, where, and to what extent conditional memory should participate in scientific reasoning. We characterize the scientific knowledge boundary and controlled interventions on memory-enabled knowledge-circuit nodes. Based on these analyses, we propose a Knowledge Boundary-Aware Router that uses task-specific input proxies available before generation to determine whether memory is activated, which layer-stage nodes receive memory signals, and how strongly these signals contribute. Experiments on biological and chemical reasoning benchmarks, covering two backbone families and six task types, show that memory effects vary substantially across inputs, tasks, and injection locations. Compared with static and activation-rate-matched random routing, our approach more consistently preserves beneficial memory contributions while suppressing memory-induced regressions, establishing selective memory allocation as an important principle for reliable scientific reasoning.
