All papers

The paper proposes an execution-gating method to safely launch and cancel tool calls early during live speech recognition before user self-corrections.

The paper proposes a graph-based task synthesis framework to help long-horizon agents satisfy inter-dependent constraints and achieve complete task outcomes.

The paper analyzes training instability cascades in large recommender systems and proposes targeted gradient, weight, and loss interventions.

Proposes a causal joint attention model across faces and time for low-latency active speaker detection in multi-person video.

This paper establishes mathematical conditions to guarantee long-term performance and low counterfactual regret for reflective language agents under resource constraints.

The paper introduces a spatial reasoning benchmark to evaluate how models handle varying reference frames and altered visual evidence.

Proposes a latent space model that summarizes historical interactions to guide decentralized multi-agent self-organization.

This paper provides a systematization of knowledge on membership inference attacks in tabular machine learning across small-to-medium datasets.

Presents an adaptive retrieval framework that selectively retrieves historical video context to assist streaming question answering.

The paper investigates how conditional flow models learn and potentially hallucinate non-additive interactions in scientific prediction tasks.

Analyzes how language models develop in-context learning by separating simple demonstration copying from multi-token rule composition across training checkpoints.

Uses task dependency graphs to help multi-agent teams maintain and evolve shared procedural memories across long tasks.

The paper evaluates how objective functions and candidate constraints affect decision conversion when generating radiology reports from sampled evidence.

The paper proposes a latent-space unlearning method that preserves model performance on benign tasks conceptually adjacent to forgotten domains.

The paper introduces a method that weights rubric criteria through pairwise model comparisons to improve evaluation accuracy.

Develops a dual-path architecture combining visual and acoustic signals to improve multimodal emotion recognition.

Introduces a latency-aware workflow learning method to reduce execution delays in recurring multi-agent queries.

The paper presents a diagnostic evaluation framework to measure how inference-time techniques affect parametric knowledge extraction in models.

The paper introduces a federated multi-agent framework that guides agent selection and query refinement under dynamic privacy constraints.

The paper introduces a reasoning-guided translation framework to scale multilingual safety and reasoning benchmarks while preserving original task meanings.

Aligns prompt semantic structures across audio and visual data to identify unseen multimodal events.

The paper introduces a framework to model and evaluate how extremist agents affect collective behavior in multi-agent language model societies.

The paper presents a framework to formalize compute, labor, and workflow costs as core measurement design criteria for automated essay scoring.

The paper uses in-context neurofeedback sessions to evaluate whether language models can intentionally control their internal activations.

Distills user aesthetic profiles to guide multimodal music models in generating personalized albums.

Proposes a graph-guided synthetic data generation framework with controlled difficulty levels for training multi-hop research agents.

The authors study why certain linearly represented facts in language models support generalization to new relations after fine-tuning while other recalled facts do not.

The paper proposes a synthetic data pipeline that extracts atomic facts from source texts to guide factual pretraining for language models.

This paper proposes a multimodal variational autoencoder framework that separates task-relevant information to improve clinical predictions even with missing modalities.

Introduces a benchmark evaluating whether language model agents can independently direct and validate goal-specified physics research investigations.

The paper detects backdoor triggers in multimodal language models by tracking output evolution trajectories across internal hidden layers.

This paper introduces a multimodal language-driven framework combining polarization and RGB cues to identify degradations and restore underwater images.

SimInter simulates conversational recommendation users by conducting synthetic interviews between language models to enrich implicit behavioral profiles.

The paper presents a fine-tuning pipeline for vision-language models to tag historical photographs using incomplete supervision labels.

The paper introduces a benchmark to evaluate how language models perform counterfactual reasoning regarding AI data center infrastructure and siting commitments.

Introduces a self-supervised consistency loss across varied tool interfaces to make language model agents robust against notation and naming changes.

The authors formulate colon cancer treatment planning as an interactive reinforcement learning task in settings where actions are only partially verifiable.

The paper analyzes how vision-language models process negation by comparing model attention mechanisms with human eye-tracking data.

The paper proposes an output-space probability interface for frozen multimodal models to improve continuous sentiment prediction using gradient-boosted trees.

The paper proposes a runtime monitoring method to detect and correct language model agents that deviate from intended procedures without reporting it.

Introduces HelpMeOut, an embodied AI benchmark for diagnosing human execution errors across household activity trajectories motivated by neuropsychology research.

Analyzes the overuse and contextual appropriateness of a common corrective phrasing pattern in Mandarin language model text.

The paper introduces a benchmark for medical knowledge poisoning and a routing controller that selectively avoids unreliable retrieved evidence during clinical reasoning.

The paper introduces an agentic framework that removes peripheral visual clues to protect geolocation privacy in images without adding visible artifacts.

The paper proposes a budgeted KV-cache repair technique to restore model performance efficiently after deleting obsolete context tokens.

This paper proposes a multi-agent system designed to reduce clinical medication errors through dynamic collaborative coordination.

This paper presents a dual-branch latent diffusion method to generate LiDAR point clouds from multi-view camera images.

The paper introduces a diagnostic framework to test whether language model agents genuinely infer opponent beliefs rather than following superficial behavioral patterns.

Introduces a controlled Minecraft benchmark to evaluate whether embodied agents acquire persistent, transferrable skills across environment resets.

This paper presents a training-free contrastive decoding framework using attention-guided perturbations to reduce hallucinations in video language models.

The paper proposes a reinforcement learning method that assigns distinct credit to initial code generation and subsequent post-execution correction steps.

Analyzes representational similarity metrics and demonstrates that input category selection has a larger impact on comparisons than the choice of metric.

Evaluates eight lightweight open-source vision-language models on optical character recognition and schema-compliant JSON extraction across heritage document collections.

The paper presents a Lean 4 runtime allowing self-extending agents to run verified computations across tools without constant human oversight.

Designs an explicit digital-twin world model where robotic agents inspect, simulate, and calibrate physical parameters before acting.

Provides a probabilistic marginal likelihood derivation explaining why heavy-tailed weight spectral metrics correlate with deep network generalization quality.

Presents parts-specific reweighting and classifier guidance to improve facial details and audio alignment in speech-driven gesture videos.

The paper proposes a likelihood-based attribution method to identify the source model of machine-generated text under distribution shifts.

The paper presents two geometric methods to evaluate where and why representational structures align across different large language models.

The paper introduces Trace-to-Architecture, a framework that diagnoses LLM agent execution failures to automatically revise agent harness components.