Tell Me What To Learn: Generalizing Neural Memory to be Controllable in Natural Language

2026-02-26

AI Summary

This work proposes a general-purpose neural memory system controlled by natural language instructions to address the limited user controllability of existing neural memory models, which struggle to adapt to heterogeneous information streams and dynamic task demands. By integrating a learnable memory architecture with an instruction parsing module, the system enables flexible, natural language–guided control over memory writing and retention—overcoming the constraints of conventional fixed-objective mechanisms. It supports lightweight, on-demand selective learning, allowing users to dictate what information is stored or preserved. Experimental results demonstrate that the proposed approach significantly mitigates catastrophic forgetting across diverse domains and substantially enhances the model’s capacity for selective learning from heterogeneous data streams.

Technology Category

• Natural Language Processing: Learning & Optimization for NLP
• Machine Learning: Neuro-Symbolic Learning
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Modern machine learning models are deployed in diverse, non-stationary environments where they must continually adapt to new tasks and evolving knowledge. Continual fine-tuning and in-context learning are costly and brittle, whereas neural memory methods promise lightweight updates with minimal forgetting. However, existing neural memory models typically assume a single fixed objective and homogeneous information streams, leaving users with no control over what the model remembers or ignores over time. To address this challenge, we propose a generalized neural memory system that performs flexible updates based on learning instructions specified in natural language. Our approach enables adaptive agents to learn selectively from heterogeneous information sources, supporting settings, such as healthcare and customer service, where fixed-objective memory updates are insufficient.

Problem

• neural memory
• controllable learning
• heterogeneous information
• natural language instruction
• continual adaptation

Innovation

• neural memory
• natural language instructions
• controllable learning
• continual adaptation
• heterogeneous information