ParamMem: Augmenting Language Agents with Parametric Reflective Memory
2026-02-26 • Machine Learning
Machine LearningMultiagent Systems
AI summaryⓘ
The authors study how language agents use self-reflection to improve their problem-solving but often get stuck repeating similar ideas. They find that having more varied reflections helps agents do better. To achieve this, they create ParamMem, a memory module that helps store and generate diverse reflection patterns. Using ParamMem, they build ParamAgent, which improves performance on tasks like coding and math by combining different types of memory. Their experiments show this approach is efficient and helps models improve themselves without needing a stronger external model.
self-reflectionlanguage agentsreflective diversityparametric memorytemperature samplingcode generationmathematical reasoningmulti-hop question answeringepisodic memorymodel transfer
Authors
Tianjun Yao, Yongqiang Chen, Yujia Zheng, Pan Li, Zhiqiang Shen, Kun Zhang
Abstract
Self-reflection enables language agents to iteratively refine solutions, yet often produces repetitive outputs that limit reasoning performance. Recent studies have attempted to address this limitation through various approaches, among which increasing reflective diversity has shown promise. Our empirical analysis reveals a strong positive correlation between reflective diversity and task success, further motivating the need for diverse reflection signals. We introduce ParamMem, a parametric memory module that encodes cross-sample reflection patterns into model parameters, enabling diverse reflection generation through temperature-controlled sampling. Building on this module, we propose ParamAgent, a reflection-based agent framework that integrates parametric memory with episodic and cross-sample memory. Extensive experiments on code generation, mathematical reasoning, and multi-hop question answering demonstrate consistent improvements over state-of-the-art baselines. Further analysis reveals that ParamMem is sample-efficient, enables weak-to-strong transfer across model scales, and supports self-improvement without reliance on stronger external model, highlighting the potential of ParamMem as an effective component for enhancing language agents.