analiza aiqbench
This arXiv preprint proposes OMCE, a contextual-bandit controller that decides when an LLM agent should use stored experience, which memory format to use, and when to consolidate new guidance. The reported gains (up to 20.5 points in first-attempt success and 5.7 to 62.5x lower token use versus ACE) are benchmark results from a single submission, so they need independent replication. For readers building agent systems, the useful takeaway is that learned routing over memory may cut inference cost, though the approach's robustness outside the tested benchmarks is still unproven.
Evolving Memory, Learning to Meta-Control: Test-Time Adaptation for Experience-augmented LLM Agents
arXiv:2611.01747v1 Announce Type: cross Abstract: Self-evolving agents accumulate experience from task streams, but existing frameworks typically rely on fixed pipelines to retrieve and utilize this experience, neglecting when and how to deploy it. We introduce Online Meta-Control of Experience (OMCE), a framework that updates experience memory and learns a meta-control policy from task feedback during test-time adaptation. OMCE employs a lightweight contextual-bandit meta-controller that uses compact retrieval signals to select among four cognitive actions: DIRECT, EPISODE, GUIDE, and DISTILL. This routing mechanism operationalizes three coupled decisions: whether to invoke experience, which memory format to utilize, and when to consolidate new guidance. OMCE evolves online, leveraging task feedback to update the memory bank and refine the routing policy via contextual bandit updates. Experiments on four online adaptation benchmarks across four backbones show that OMCE improves first-attempt success by up to 20.5 percentage points over the strongest compared baseline, while reducing per-task token use by a factor of 5.7 to 62.5 relative to ACE with GPT-5.2. Controlled ablations further support the benefits of learned routing and continued routing policy adaptation.
Abstract:Self-evolving agents accumulate experience from task streams, but existing frameworks typically rely on fixed pipelines to retrieve and utilize this experience, neglecting when and how to deploy it. We introduce Online Meta-Control of Experience (OMCE), a framework that updates experience memory and learns a meta-control policy from task feedback during test-time adaptation. OMCE employs a lightweight contextual-bandit meta-controller that uses compact retrieval signals to select among four cognitive actions: DIRECT, EPISODE, GUIDE, and DISTILL. This routing mechanism operationalizes three coupled decisions: whether to invoke experience, which memory format to utilize, and when to consolidate new guidance. OMCE evolves online, leveraging task feedback to update the memory bank and refine the routing policy via contextual bandit updates. Experiments on four online adaptation benchmarks across four backbones show that OMCE improves first-attempt success by up to 20.5 percentage points over the strongest compared baseline, while reducing per-task token use by a factor of 5.7 to 62.5 relative to ACE with GPT-5.2. Controlled ablations further support the benefits of learned routing and continued routing policy adaptation.
23 pages, 5 figures. Submitted to ACL Rolling Review (October 2026)
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2611.01747 [cs.LG]
(or arXiv:2611.01747v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2611.01747 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 285 cuvinte