arXiv:2607. 21111v1 Announce Type: cross Abstract: Offline Reinforcement Learning (RL) agents are trained on fixed behavioral trajectories, which makes trajectory-level deletion important when selected data must be removed after training.
arXiv:2607. 28829v1 Announce Type: cross Abstract: Self-improving federated agent networks keep training after deployment by collecting new trajectories with the current policy and feeding them back into later rounds.
arXiv:2608.20784v1 Announce Type: cross Abstract: Imitation learning for robotics depends on human demonstrations, some of which people may later ask to remove. Retraining without them is the natural...
arXiv:2607. 09773v1 Announce Type: new Abstract: Computer-use agents must solve long-horizon tasks through repeated interaction with partially observable, multimodal desktop environments.
Long-horizon language-model agents accumulate reasoning traces, tool exchanges, and observations whose relevance changes with the current decision. Existing compression strategies often score historic...
arXiv:2609.05837v1 Announce Type: new Abstract: LLM-based agents are increasingly deployed in real-world applications through tool-use APIs, yet training them for specific environments remains fundam...
The paper introduces Direct Relational Set‑Risk Pruning (DRSR), a method for compressing the history of long‑horizon language‑model agents by selecting deletion sets based on risk constraints rather than independent unit scores. DRSR builds counterfactual supervision offline, then uses a lightweight scorer to predict set‑level harm during deployment, removing the largest safe set while respecting recency, protocol, and budget limits. Experiments on WorkBuddyBench Full260 and Eval40 show that DRSR improves mean reward from 0.699 to 0.802 and reduces token usage by over 20%, with further analyses highlighting the importance of decision‑conditioned relations, retained context, pair interactions, and abstention.
arXiv:2602. 05459v2 Announce Type: replace Abstract: Offline goal-conditioned reinforcement learning (GCRL) is typically benchmarked by the best tuned success rate of each method.
arXiv:2605. 24202v2 Announce Type: replace Abstract: Multi-agent LLM workflows route inference through specialized roles to lift end-task accuracy, but jointly training those roles with reinforcement learning is unstable in ways that are poorly understood.
arXiv:2606. 10062v1 Announce Type: new Abstract: Foundation-model agents are increasingly long-lived systems that remember users across interactions, making memorization an explicit deployment-time function rather than solely a property of model weights.
PEARL is a framework for human‑centric cyber‑physical systems that uses a dual‑path Early‑Exit Deep Q‑Network to control the trade‑off between privacy and utility. By training per‑branch binary labels—Utility Confidence Labels (UCL) and Privacy Confidence Labels (PCL)—based on mutual information between private states and observable actions, PEARL selects the shallowest exit that satisfies both privacy and utility constraints, avoiding noise injection. The system includes an MI‑based feedback loop to detect behavioral drift and trigger retraining, and experiments on a smart‑home HVAC system and a VR smart classroom show a 25.67% reduction in adversarial state‑inference accuracy with only a 10‑16% utility cost.
arXiv:2601. 07376v2 Announce Type: replace Abstract: We introduce \textsc{OpenTinker}, an open infrastructure for training large language model (LLM) agents with many LoRA-backed policies over shared execution resources.
arXiv:2609. 04875v1 Announce Type: cross Abstract: Long-running LLM agents are stateful: beyond the transcript they accrete compressed summaries, plaintext memory, pending tool plans, and, under every serving API, a KV cache.