Problem
Research questions and friction points this paper is trying to address.
on-device LLM agents
continual learning
memory budget
poisoning
experience memory
Forget to Improve: On-Device LLM-Agent Continual Learning via Budget-Curated Memory
🤖 AI Summary
This work addresses the security risks—such as data poisoning—and resource inefficiencies arising from uncontrolled memory accumulation in on-device large language model agents operating under stringent constraints on memory, energy, and uplink bandwidth. To this end, we propose the first unified memory governance framework that jointly evaluates both the utility and potential harm of stored memories. Constrained by explicit resource budgets, the framework introduces a per-byte net-value metric (utility minus harm) to simultaneously optimize decisions on whether to retain (KEEP), share (SHARE), or trust (TRUST) each memory entry. Experiments on Jetson heterogeneous platforms demonstrate that our approach reduces memory footprint by 2.7× and uplink communication by 2.4× compared to baselines, completely eliminates successful injection attacks (reducing success rate from 0.75 to zero), and significantly improves task accuracy under interference from poisoned or stale memories.
📝 Abstract
On-device language-model agents improve by accumulating experience in retrieved memory rather than by updating weights. This memory is hard-bounded and exposed: it consumes RAM and energy, reaches peers through a thin uplink, and becomes an attack surface because it is writable by what the agent reads. Existing systems each cover one part of this problem: agentic memories grow without a budget, on-device methods keep entries by success alone, and poisoning is studied mainly as an attack rather than as a memory-governance problem. We propose \sys{}, a single net-value-per-byte score that governs an agent's experience-memory lifecycle. The main idea is to let the budget act as the curator: each entry is scored as value minus harm, per byte, so one ruler decides what to keep, share, and trust. \sys{} makes three decisions: (1) \textbf{KEEP} evicts low-value bytes under the RAM and energy budget; (2) \textbf{SHARE} sends an insight only when its value exceeds its uplink cost; and (3) \textbf{TRUST} gates a peer entry by provenance. On language-model-agent task-drift benchmarks and a real heterogeneous Jetson testbed with two robot-arm nodes and a hub, \sys{} reduces memory by $2.7\times$ and uplink by $2.4\times$, drives injection success from 0.75 to zero, and raises accuracy on cases corrupted by poison or stale memory. Curating by net value reduces footprint, energy, uplink, and injection success together without reducing accuracy. In this setting, forgetting by net value improves the agent rather than weakening it.
Innovation
Methods, ideas, or system contributions that make the work stand out.
continual learning
on-device LLM agents
memory curation
net-value-per-byte
poisoning defense
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
