CASCADE: Case-Based Continual Adaptation for Large Language Models During Deployment

2026-05-05

AI Summary

This work addresses the limited continual learning capability of large language models after deployment, which hinders their ability to evolve through environmental interaction. The authors propose Deployment-Time Learning (DTL) as a third phase in the model lifecycle, introducing an explicit, evolvable episodic memory system that enables ongoing adaptation without updating model parameters. DTL formalizes the deployment stage as a continual learning process for the first time, integrating case-based memory with a contextual multi-armed bandit algorithm to achieve theoretically grounded exploration–exploitation trade-offs and long-term no-regret performance. Evaluated across 16 diverse tasks, the approach improves macro-average success rate by 20.9% over zero-shot prompting and significantly outperforms both gradient-based and memory-based baselines.

Abstract

Large language models (LLMs) have become a central foundation of modern artificial intelligence, yet their lifecycle remains constrained by a rigid separation between training and deployment, after which learning effectively ceases. This limitation contrasts with natural intelligence, which continually adapts through interaction with its environment. In this paper, we formalise deployment-time learning (DTL) as the third stage in the LLM lifecycle that enables LLM agents to improve from experience during deployment without modifying model parameters. We present CASCADE (CASe-based Continual Adaptation during DEployment), a general and principled framework that equips LLM agents with an explicit, evolving episodic memory. CASCADE formulates experience reuse as a contextual bandit problem, enabling principled exploration-exploitation trade-offs and establishing no-regret guarantees over long-term interactions. This design allows agents to accumulate, select, and refine task-relevant cases, transforming past experience into actionable knowledge. Across 16 diverse tasks spanning medical diagnosis, legal analysis, code generation, web search, tool use, and embodied interaction, CASCADE improves macro-averaged success rate by 20.9% over zero-shot prompting while consistently outperforming gradient-based and memory-based baselines. By reframing deployment as an adaptive learning process, this work establishes a foundation for continually improving AI systems.

Problem

• deployment-time learning
• continual adaptation
• large language models
• episodic memory
• lifecycle

Innovation

• deployment-time learning
• case-based adaptation
• episodic memory
• contextual bandits
• continual learning