All papers

Presents PolyGraph, a method storing provenance receipts during knowledge graph construction to enable precise updates when source documents are retracted.

Benchmarks combinations of language models, communication protocols, and agent frameworks on web interaction tasks.

Proposes Rollout-Time Intervention to convert failed agent trajectories into successful off-policy training signals during reinforcement learning.

Proposes SWE-Critix, a post-training pipeline designed to train verifiers that evaluate and explain long coding-agent execution trajectories.

Presents Agent4Spec, a multi-agent framework that automatically validates OpenAPI syntax errors and identifies functional API gaps through market analysis.

Presents MENTOR, a multi-agent framework that intervenes adaptively during failed reasoning rollouts to prevent zero-reward collapse in RLVR.

Presents RESCUE, a counterfactual simulation framework to evaluate whether runtime failure monitor alarms enable beneficial robot interventions.

Co-trains an interactive agent's task policy with a clarification-value controller to optimize when to ask follow-up questions.

Develops a collaborative learning framework that splits large language models into smaller heterogeneous slices to mitigate error propagation.

Presents a framework that dynamically adjusts role activation and multi-agent collaboration based on the model backbone's query-specific competence.

Introduces a benchmark evaluating the risks of LLM agents designing dangerous biological mutations when augmented with biodesign tools.

Presents an agent harness that coevolves evidence and conclusions to improve reasoning across 100-million-token document corpora.

Formulates a population dynamics model to study takeoff thresholds and self-replication risks in collaborative AI agents.

The paper proposes a framework where an agent policy navigates a global search tree by alternating between reflection and execution roles.

Proposes a streaming optimal transport method to monitor LLM agent trajectories in real time against reference runs.

Presents a multi-agent self-supervision method that uses evolutionary workflow optimization and fine-tuning for recursive model self-improvement.

Proposes a training method to adaptively skip turn-by-turn agent reasoning when previous thoughts remain sufficient for future actions.

Presents ReSI, a recursive safety framework using cyclical evaluation and updates to improve AI resilience against emergent threats.

Presents a simulated participant environment to evaluate multi-turn AI interviewers across market research and social science scenarios.

Surveys and taxonomizes recurring architectural patterns across language-model-integrated software applications, ranging from chatbots to autonomous agent systems.

Proposes a method linking execution traces to specific skill segments to guide automated revision of software engineering agent skills.

Proposes an external memory framework allowing frozen language model agents to learn and iteratively refine natural-language world models compiled into executable code.

Proposes a narrow-and-deep ontology tower to structure equipment knowledge for language model agents operating industrial air-handling systems.

Proposes study contracts to register execution obligations and verify research agent outputs against approved protocol parameters.

Presents a portfolio management agent that uses fractional factorial counterfactual evaluation and Banzhaf values to isolate memory utility from market movements.

Introduces a step-level memory reuse framework using transition graphs to adapt relevant prior mobile graphical interface execution trajectories.

Presents a framework that defers memory construction to test time to answer queries across long interaction histories.

Proposes a framework that encodes visual trajectories into reusable skill cards to reinforce vision-language agent planning and rewards.

Develops a framework for egocentric multimodal assistants to decide both when to speak and what guidance to provide.

Introduces a preview-guided admission framework allowing search agents to selectively incorporate retrieved candidates into the ongoing reasoning context.

Presents a temporal memory architecture using hierarchical tree structures to trace evolving multi-party conversational histories for agents.

Generates synthetic multi-agent dialogues to benchmark conversational agent robustness against non-cooperative and adversarial user behaviors.

The paper introduces a benchmark evaluating how multi-agent teams develop communication protocols under asymmetric information and restricted communication channels.

The paper presents a framework to probabilistically localize user-intent misinterpretations and repair conversational states in multi-turn assistant systems.

The paper investigates how observing partner states, sharing knowledge, and predicting partner actions affect coordination in embodied multi-agent teams.

Applies regenerative software primitives so programs can be deleted and rebuilt by agents without losing architectural intent and knowledge.

The paper introduces an optimization method that separates error identification from credit assignment to improve reinforcement learning for tool-using agents.

Introduces a closed-loop benchmark testing language models controlling physical robotic hardware and reacting to real sensor feedback.

The paper introduces a neuro-symbolic benchmark to evaluate whether language model agents actively investigate environments and make causal decisions.

ScaleMAS coordinates multi-agent systems using an asynchronous shared datastore with limited search to improve scaling and fault tolerance.

The paper presents SPROUT, a snapshot-based rollout framework to scale reinforcement learning for long-horizon software engineering tasks.

This paper introduces learned communication adapters and action heads to coordinate multiple vision-language-action robotic policies during collaborative tasks.

Applies inference-time activation steering to keep multi-agent systems from overstepping their assigned roles.

Evaluates whether failure feedback helps self-improving agents fix specific target bugs rather than general errors.

Introduces multi-step action forecasting targets alongside actor rollouts to train language agents more effectively from intermediate trajectory steps.

The paper presents a deterministic context compaction method designed to preserve factual history and commitments in lightweight coding agents.

The paper introduces an automated pipeline that builds repository-level software engineering benchmarks from test execution footprints.

The paper introduces a benchmark to evaluate how personal assistants handle interdependent personal and procedural memories over long trajectories.

This paper develops a runtime verification architecture that decouples monitoring formalisms from execution boundaries to safely control AI agents.

The paper analyzes prefix replay curricula and handoff rules under reinforcement learning to improve long-horizon task completion for language model agents.

The paper presents an end-to-end trained multimodal agent that combines visual graphical interface actions and executable code execution.

This paper introduces a framework for self-improving agents that records and shares both failure and success cases to prevent redundant mistakes.

Presents a training-free framework that tracks changing conversational facts as versioned records to improve agent memory retrieval.

The paper introduces action chunking for agentic robotics to execute pre-planned primitive routines without redundant model reasoning at each step.

The paper introduces an auditing framework that identifies failure steps in multi-agent systems by evaluating how information is utilized across handoffs.

The authors introduce a benchmark to evaluate autonomous scientific discovery agents making iterative experimental decisions under fixed budgets.

AssetForge pairs reusable application assets with task rubrics to synthesize executable training tasks and evaluation checks for business automation agents.

The authors propose verifiable agent insurance using smart-contract escrow to financially compensate users when agents violate programmatically logged behavioral promises.

TextualRL updates agent context prompts using groupwise trajectory comparisons rather than isolated feedback runs to improve policy adaptation without parameter tuning.

This paper presents an evaluation benchmark and guardrail framework for detecting safety risks in multi-turn interactions with AI companions.