All papers

The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models.

The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features.

The paper proposes a decoupled per-token scoring method to efficiently manage key-value cache memory during long-context generation.

The paper coordinates projector and language model adaptation using task-specific routing to reduce catastrophic forgetting in multimodal models.

The paper uses retrieved in-context exemplars to improve rhetorical structure discourse parsing without requiring model fine-tuning.

The paper introduces a benchmark to measure language model agent performance under real-time latency and streaming temporal constraints.

The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time.

The paper develops a multilingual retrieval-augmented tutoring system providing bilingual responses and glossaries for international computing students.

The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions.

The paper trains feedforward 3D reconstruction models using privileged motion and gravity cues to improve dynamic scene understanding.

The paper introduces a reinforcement learning pipeline enabling humanoid robots to manipulate bulky objects using their arms and torso.

The paper proposes an iterative self-improvement method combining trajectory rewriting and policy refinement for terminal-based coding agents.

The paper introduces a multi-agent system to automate data curation and dataset editing for supervised fine-tuning.

The paper introduces a hierarchical task network framework that separates multi-agent planning from execution scheduling.

The paper examines decentralized multi-agent coordination through sequential market auctions under conditions of imperfect competition.

The paper introduces an adaptive connectivity objective to improve deep learning segmentation of thin, continuous anatomical or physical structures.

The paper proposes a graph neural network method to speed up layout decomposition for advanced semiconductor manufacturing.

The paper introduces a skill-decoupled on-policy distillation method to merge vision-language experts without conflicting gradients or reward hacking.

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

The paper presents a closed-form, training-free approach to merge specialized fine-tuned models into a single student using forward passes.

The paper develops an unsupervised routing method to escalate tasks from weak language models to stronger models without ground-truth labels.

The paper introduces an agentic reasoning controller framework and an evaluation benchmark for optimizing open radio access networks.

The paper trains language model agents to reorganize unstructured document collections into structured workspaces to speed up future multi-document queries.

The authors evaluate how developer scaffolding and general prompt instructions affect the performance of environment-agnostic coding agents across diverse benchmarks.

The authors introduce EliteReplay, combining reinforcement learning with quality-diversity search replay to retain solution diversity during post-training.

The authors propose Subspace Tuning, a fine-tuning method restricting weight updates to bottom singular subspaces to mitigate catastrophic forgetting in language models.

The authors develop a clinical benchmark evaluating language model pipelines for generating multidisciplinary second opinions on complex medical cases.

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 authors report on a team of autonomous Gemini 4 Argon agents collaborating via chat and shared code to beat Minecraft survival mode.

The authors present AVC-Reward, an event-centric framework to evaluate temporal, narrative, and attribute consistency in generated audio-video content.

The authors present a market-based post-training framework that allows decentralized agent swarms to learn both action policies and coordination mechanisms without central orchestration.

The authors propose combining line plots and normalized delay embeddings to provide complementary visual representations for time-series reasoning in language models.

The authors propose ACEdit, a training and distillation scheme that stabilizes image editing models against accumulating errors across multiple editing turns.

The authors introduce LangDiT, an adapted transformer architecture designed specifically for continuous diffusion language models.

The authors propose KnowGPO, a reinforcement learning framework that assigns credit to deep search agents based on acquired knowledge conditions rather than trajectories.

The authors introduce DSP, a framework where robots learn from brief human play and autonomously collect contact-rich data for dexterous manipulation.

The authors introduce 3DCodeVerse, a dataset and multi-agent framework enabling coding agents to generate, reason over, and verify 3D executable code.

The authors introduce GeoState, a view-decoupled 3D world modeling approach separating scene dynamics from camera motion.

The authors propose a post-training method enabling autonomous language model agents to organize experiments and carry out machine learning engineering tasks.

The authors introduce FDE-Gym, a benchmark of 95 customer delivery tasks evaluating AI agents on ambiguous requirements and live software environments.

The authors propose MirrorMoE to reconcile an exploration dilemma between fine-grained expert segmentation and shared experts in mixture-of-experts models.

The authors present CoDA, a framework where a simulation data engine and a robot action policy co-evolve using memory of past failures.

The authors propose T2-Z, a method optimizing learnable environment latents at test time to help robotic world models adapt to unobserved physical factors.

The authors introduce an evolutionary framework using large language models to design and optimize resource-efficient quantum circuits.

The authors introduce LIBERO-PeRM, a simulation benchmark for studying how embodied robotic agents learn and follow user preferences in manipulation tasks.

The paper evaluates twenty-one language models used as judges across three benchmarks, finding widespread discrepancies between agreement metrics and underlying validity.

The paper analyzes extreme faults in modular multilevel converters and proposes a protection scheme using integrated gate-commutated thyristors.

The paper presents a network utilizing self-attention and spatio-temporal convolutions for 3D skeleton-based human action recognition.

The paper proposes an uncertainty-aware multimodal descriptor fusing sequential camera and LiDAR data for robotic place recognition.

The paper develops a microprocessor-based smart wearable glass system with GPS monitoring to assist visually impaired individuals.

The authors introduce CharXiv, a benchmark of realistic scientific charts designed to test descriptive and reasoning capabilities in multimodal language models.

This paper analyzes deprecated Android application programming interfaces and derives patterns to study how developers update them.

This paper profiles the maintainability and structural code quality of model-generated Python programs using static analysis tools.

This paper introduces a benchmark to evaluate software engineering language agents on migrating code between Linux and Windows.

This paper proposes a framework to detect training data contamination in bug-fixing benchmarks for language models.

This paper introduces an approach to detect statistically significant software energy consumption regressions across commits in version histories.

This paper develops machine learning models to detect, explain, and suggest alternatives for offensive language in online developer communities.

This paper presents a fine-tuning method to update language models for code comprehension as codebases change over time.

The paper extracts software requirements and architecture from code repositories and tests whether language agents can reconstruct projects from them.

The paper models mixed human-agent software teams to find cost-optimal spending allocations balancing human developer hours and automated coding agents.