All papers

The paper learns trajectory embeddings to predict the success or failure of language model agents early during execution.

The paper analyzes the impact of teacher rationales before distilling medical reasoning capabilities into compact language models.

This paper presents a benchmark to evaluate performance, resource usage, and privacy trade-offs when language model agents collaborate across edge and cloud environments.

Introduces ELMA, a benchmark dataset for generating long-term human motions from multi-segment instructions with anaphoric references.

Describes an industrial framework for evaluating, ablating, and deprecating features in production recommender systems without performance regressions.

This paper introduces a benchmark evaluating whether coding agents can reimplement and train biological AI tools from published scientific papers.

Proposes a knowledge distillation method that focuses learning signals on specific structural disagreements between teacher and student representations.

This paper presents a compact decision model designed to screen malicious agent tool skills with high recall directly from specification files.

Anchors feed-forward visual geometry to LiDAR-inertial motion to produce metrically accurate trajectories and 3D Gaussian reconstructions.

Examines error localization and credit assignment failures in critique-conditioned on-policy self-distillation for mathematical reasoning.

This paper introduces a multimodal benchmark evaluating AI models on evidence-grounded decision making in chemistry and materials laboratories.

This paper presents a structured pruning framework that decouples data-free neuron selection from localized weight reconstruction to reduce model memory requirements.

Introduces Socratic Skill Learning, interleaving expert guidance with student practice to produce reusable and transferable agent skills across tasks.

Presents a previsualization workflow combining physical video capture and generative reshooting to evaluate and manage visual hallucination risk.

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

The paper introduces an evaluation framework that diagnoses bottlenecks and dynamically regenerates benchmark tasks for language model agents.

The paper distills an LLM-based search relevance judge into smaller multimodal student models for efficient e-commerce search evaluation.

Presents ADMA, an adaptive deliberation controller that dynamically adjusts multi-agent critique and coordination to improve reasoning accuracy.

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

Introduces a Ukrainian meme dataset and evaluates vision-language models on detecting persuasive and propaganda techniques.

The paper introduces a multilingual benchmark to evaluate how language models identify financial concepts and extract supporting numerical evidence.

This paper introduces an aligned language model family trained with structured reinforcement learning for clinical reasoning and traditional Chinese medicine prescribing.

LatentRemedy uses task-conditioned latent tokens to allow safe key-value cache reuse across conflicting tasks in multi-agent systems.

Introduces an annotated dataset of handwritten student answer sheets with complex formulas, text, tables, and noise artifacts.

Investigates multimodal LLM failures on squint-readable low-frequency text and provides a diagnostic benchmark alongside training interventions.

The paper presents an evaluation framework to test the interactive agentic capabilities of base language models before post-training.

The authors introduce an on-policy distillation method to transfer knowledge from autoregressive language models into diffusion language models using autoregressive rollouts.

This paper introduces a multi-sensor, multi-temporal satellite imagery dataset designed for pretraining Earth observation foundation models.

The paper introduces local contract-conditioned rewriting to prevent sensitive data leakage in email-generating language model agents cost-effectively.

Proposes an inference-time unlearning method that detects leaks and injects corrective embeddings into large reasoning model traces.

The paper introduces an internal probing method to detect whether safety-aligned language models still retain accessible representations of hazardous concepts.

This paper presents an integer-only local learning system on edge FPGAs for energy-efficient convolutional neural network training.

The authors examine why language model agents fail by analyzing discrepancies between internal representations of state values and actual action choices.

The authors introduce a benchmark evaluating egocentric video understanding structured around an interactive loop of user perception, action, and world feedback.

Demonstrates a backdoor attack where malicious reasoning degradation is triggered incrementally using invisible prompt tokens and stage tracking.

The paper analyzes Best-of-N sampling as an extreme-value competition between tails of correct and incorrect responses to explain reward hacking.

The paper introduces structured random transform constructions to accelerate 1-bit quantized Johnson-Lindenstrauss compression for key-value caches.

The authors present a parallel decoding method that speeds up on-device tool calling for language model agents without requiring model retraining.

The paper introduces a dataset and benchmark to evaluate vision-language models on recognizing and answering questions about Philippine cultural heritage sites.

Analyzes layer-wise visual interventions in multimodal language models to identify where visual edits effectively reduce hallucinations.

Introduces a benchmark dataset and baseline architecture for reconstructing standalone 3D garments from single images.

The paper evaluates the reliability, score variance, and consistency of automated peer reviews generated by large language models.

The paper investigates methods for encoding context into model weights so that language models properly apply the temporary context during subsequent inference.

The paper presents a unified node-centric framework to quantify and manage multi-agent system lifecycles across design, runtime, and post-hoc attribution.

Proposes a memory-guided differential privacy framework enabling private inference over million-token contexts in LLM-as-a-service settings.

Introduces a reinforcement learning objective that groups prompt tokens into semantic spans to refine text-to-image prompt rewriting.

The paper introduces a framework to audit cultural, geographic, and user-representation biases embedded within popular machine learning benchmarks.

The paper proposes a generation-aware singular value decomposition method to compress large language models while preserving generative output quality.

The paper proposes a training-free framework that co-evolves an executable verifier program and harness state to self-improve agents on open-ended tasks.

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

The paper introduces an evergreen benchmark to evaluate the ability of language models to retrieve and format multi-element factual lists.

The paper provides an analytical framework for how multimodal agents form, represent, and adapt memory over long-horizon tasks.

Introduces bidirectional interleaved pipeline parallelism to improve throughput and activation memory scaling in long-sequence model training.

The paper proposes a system for physical AI that continuously monitors environments to identify intervention needs and execute timely actions.

Investigates internal transformer mechanisms, showing attention updates decode harmfulness while MLP layers primarily drive refusal token generation.

This paper introduces ChronoCulture to evaluate how well language models capture temporal changes in cultural attitudes across 72 countries.

This paper introduces a scalar quantization method using analytic coding to preserve language model predictive fidelity at precisions down to two bits.

The paper presents an efficient video super-resolution method using hybrid attention and reward-yield distribution matching distillation.

Formulates machine unlearning for LLM agents across parametric memory, external persistent storage, and tool-accessible environments.

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