Beyond Training: Enabling Self-Evolution of Agents with MOBIMEM

2025-12-15

AI Summary

Existing LLM-based agents lack post-deployment self-evolution, relying heavily on frequent retraining—entailing prohibitive computational costs—and face an inherent trade-off between accuracy and reasoning efficiency. This paper proposes MOBIMEM, a memory-centric agent system that decouples agent evolution from model weights via three specialized, modular memory primitives: Profile, Experience, and Action. It integrates OS-level services—including a lightweight scheduler, action logging/replay, and context-aware anomaly recovery—to enable safe, autonomous evolution. Key innovations include DisGraph indexing for efficient memory retrieval, multi-level templated execution logic, fine-grained action sequence modeling, and the AgentRR mechanism for robust runtime adaptation. Experiments show MOBIMEM achieves 83.1% Profile alignment, reduces retrieval latency to 23.83 ms (280× faster than GraphRAG), improves task success rate by up to 50.3%, and cuts end-to-end latency to one-ninth of prior approaches.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Modeling other Agents
• Machine Learning: Mixture of Experts (MoE)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Model (LLM) agents are increasingly deployed to automate complex workflows in mobile and desktop environments. However, current model-centric agent architectures struggle to self-evolve post-deployment: improving personalization, capability, and efficiency typically requires continuous model retraining/fine-tuning, which incurs prohibitive computational overheads and suffers from an inherent trade-off between model accuracy and inference efficiency. To enable iterative self-evolution without model retraining, we propose MOBIMEM, a memory-centric agent system. MOBIMEM first introduces three specialized memory primitives to decouple agent evolution from model weights: (1) Profile Memory uses a lightweight distance-graph (DisGraph) structure to align with user preferences, resolving the accuracy-latency trade-off in user profile retrieval; (2) Experience Memory employs multi-level templates to instantiate execution logic for new tasks, ensuring capability generalization; and (3) Action Memory records fine-grained interaction sequences, reducing the reliance on expensive model inference. Building upon this memory architecture, MOBIMEM further integrates a suite of OS-inspired services to orchestrate execution: a scheduler that coordinates parallel sub-task execution and memory operations; an agent record-and-replay (AgentRR) mechanism that enables safe and efficient action reuse; and a context-aware exception handling that ensures graceful recovery from user interruptions and runtime errors. Evaluation on AndroidWorld and top-50 apps shows that MOBIMEM achieves 83.1% profile alignment with 23.83 ms retrieval time (280x faster than GraphRAG baselines), improves task success rates by up to 50.3%, and reduces end-to-end latency by up to 9x on mobile devices.

Problem

• Enables self-evolving agents without continuous model retraining
• Decouples agent evolution from model weights using memory primitives
• Improves personalization, capability, and efficiency on mobile devices

Innovation

• Memory-centric system decouples agent evolution from model weights
• Specialized memory primitives optimize profile retrieval and task execution
• OS-inspired services orchestrate execution with scheduling and exception handling