Forget to Improve: On-Device LLM-Agent Continual Learning via Budget-Curated Memory

• URL: http://arxiv.org/abs/2606.25115v1
• Date: Tue, 23 Jun 2026 19:42:07 GMT
• Title: Forget to Improve: On-Device LLM-Agent Continual Learning via Budget-Curated Memory
• Abstract summary: On-device language-model agents improve by accumulating experience in retrieved memory rather than by updating weights.<n>We propose sys, a single net-value-per-byte score that governs an agent's experience-memory lifecycle.
• Score: 7.254575731198877
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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.

Related papers

• The Compliance Trap: Diagnosing How AI Agents Consume Conflicting Memory [9.041305193333434]
  Memory is becoming a core component of long-horizon AI agents, allowing agents to reuse past experience when operating web browsers, software tools, and other interactive environments.<n>Existing work mostly treats memory as a supply problem, asking what experience to write, how to store it, and which entry to retrieve for the next task.<n>Yet we still lack a clear account of how models consume retrieved memory across a multi-step action trajectory.
  arXiv  Detail & Related papers  (2026-07-12T07:05:31Z)
• Memory as a Wasting Asset: Pricing Flash Endurance for Embodied Agents, and the Limits of Doing So [0.0]
  A robot's flash endurance is a non-renewable stock, yet no fielded robot memory system prices which memories are worth an erase cycle.<n>We treat embodied memory as depreciating capital and price that stock with a single endurance shadow price $$, which makes cost-minimizing placement across a RAM / on-board NVM / cloud hierarchy a threshold in a wear-augmented per-byte index.
  arXiv  Detail & Related papers  (2026-06-16T16:43:19Z)
• MemVenom: Triggered Poisoning of Multimodal Memories in Web Agents [74.64265314956441]
  We propose a unified black-box attack framework that poisons graph-structured external memory with coordinated text-image evidence.<n>MemVenom achieves strong end-to-end attack success with minimal impact on benign performance, reaching up to 99.15% on GPT-5-family web agents.
  arXiv  Detail & Related papers  (2026-06-09T11:53:25Z)
• Enhancing Software Engineering Through Closed-Loop Memory Optimization [50.418699819003486]
  We introduce ours, a closed-loop framework for memory augmentation in software engineering (SE) agents.<n>ours grounds memory utility in textitvalidated downstream impact, establishing utility as a task-agnostic textbf benchmark and an annotation-free textbfoptimization signal.<n>Results demonstrate that ours consistently improves SE agents across settings, achieving absolute gains of up to $uparrow5.25%$ in success rate and $uparrow4.63%$ in resolve efficiency.
  arXiv  Detail & Related papers  (2026-06-04T03:17:21Z)
• WorldMemArena: Evaluating Multimodal Agent Memory Through Action-World Interaction [72.1620416874118]
  Multimodal large language models are increasingly deployed as long-horizon agents.<n>Existing benchmarks measure recall over static dialogue, collapse memory into a single end-of-task accuracy, and reduce visual observations to captions.<n>We formulate multimodal agent memory as an Action-World Interaction Loop with an observable four-stage lifecycle.
  arXiv  Detail & Related papers  (2026-05-28T04:27:20Z)
• MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection [15.403262505444454]
  We propose a post-hoc causal memory auditing framework for large language model agents.<n>MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios.
  arXiv  Detail & Related papers  (2026-05-22T15:03:13Z)
• MemGym: a Long-Horizon Memory Environment for LLM Agents [69.79226770543049]
  We present MemGym, a benchmark for agentic memory that unifies existing agent gyms and in-house memory-grounded pipelines.<n>MemGym reports memory-isolated scores that decouple memory performance from reasoning, retrieval, and tool-use ability.<n>Our synthetic pipelines for MEMGYM-CODEQA and MEMGYM-DR are length-controllable, ablation-verified at every stage, and tightly aligned with downstream scenarios.
  arXiv  Detail & Related papers  (2026-05-20T07:25:33Z)
• PREPING: Building Agent Memory without Tasks [65.75870568959644]
  Preping is a proposer-guided memory construction framework.<n>At its core is proposer memory, a structured control state that shapes future practice.<n>Trials show Preping substantially improves over a no-memory baseline.
  arXiv  Detail & Related papers  (2026-05-11T04:34:43Z)
• Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution [52.76038908826961]
  We propose $textbfReMe$ ($textitRemember Me, Refine Me$) to bridge the gap between static storage and dynamic reasoning.<n>ReMe innovates across the memory lifecycle via three mechanisms: $textitmulti-faceted distillation$, which extracts fine-grained experiences.<n>Experiments on BFCL-V3 and AppWorld demonstrate that ReMe establishes a new state-of-the-art in agent memory system.
  arXiv  Detail & Related papers  (2025-12-11T14:40:01Z)
• A-MemGuard: A Proactive Defense Framework for LLM-Based Agent Memory [31.673865459672285]
  Large Language Model (LLM) agents use memory to learn from past interactions.<n>An adversary can inject seemingly harmless records into an agent's memory to manipulate its future behavior.<n>A-MemGuard is the first proactive defense framework for LLM agent memory.
  arXiv  Detail & Related papers  (2025-09-29T16:04:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.