9546 papers
This paper reveals that SpeechLLM backbones struggle with conversational disfluencies due to a bias toward semantic abstraction over structural fidelity, with performance varying by architecture and fine-tuning often compromising generalization despite achieving state-of-the-art results.
This paper introduces Complexity-Regularized Proximal Policy Optimization (CR-PPO), a novel algorithm that replaces standard entropy regularization with a self-regulating complexity term—defined as the product of Shannon entropy and disequilibrium—to maintain beneficial stochasticity while reducing sensitivity to hyperparameter tuning and avoiding the overriding of reward signals.
BridgeDrive introduces a novel anchor-guided diffusion bridge policy that ensures theoretical consistency between forward and reverse processes to transform coarse expert trajectories into refined, safe, and reactive closed-loop plans, achieving state-of-the-art performance on the Bench2Drive benchmark.
This paper investigates the phenomenon of subliminal learning in language model distillation, revealing that hidden bias transfer occurs under hard distillation not through global entanglement, but via a small set of critical "divergence tokens" processed in early layers, making the effect both mechanistically specific and fragile to prompt variations.
BeyondBench introduces a contamination-resistant evaluation framework that uses on-the-fly algorithmic problem generation to assess the true reasoning capabilities of 101 language models across 44 tasks, revealing significant performance gaps in complex problem-solving and the critical role of tool usage.
This paper introduces a novel NVFP4 training framework that combines Random Hadamard transforms, 2D quantization, stochastic rounding, and selective high-precision layers to successfully pretrain a 12-billion-parameter model on 10 trillion tokens with performance comparable to FP8 baselines, thereby demonstrating the viability of stable 4-bit precision training for large language models.
This paper introduces PrefDisco, a novel benchmarking framework and evaluation methodology that transforms static tasks into interactive scenarios to assess and improve large language models' ability to proactively identify user preferences and adapt their reasoning chains for personalized alignment in just-in-time, cold-start situations.
This paper introduces EgoTraj-Bench, the first real-world benchmark for ego-centric trajectory prediction under noisy observations, and proposes BiFlow, a dual-stream flow matching model with an EgoAnchor mechanism that achieves state-of-the-art robustness by concurrently denoising historical inputs and forecasting future motion.
Graph2Eval is a knowledge-graph-driven framework that automatically generates scalable, semantically consistent, and solvable multimodal agent tasks by leveraging structured subgraph sampling and multi-stage filtering, resulting in the Graph2Eval-Bench dataset which significantly improves evaluation reliability over existing baselines.
This paper introduces SpineMed, a clinician-co-designed ecosystem featuring the 450k-instance SpineMed-450k dataset and the SpineBench evaluation framework, which together enable and validate significant improvements in level-aware, multimodal reasoning for spine disorder diagnosis and surgical planning.
MachaGrasp is an eigengrasp-based, end-to-end framework that generates dexterous grasp articulations across different hand embodiments by leveraging morphology embeddings and a kinematic-aware loss, achieving high success rates in both simulation and real-world few-shot adaptation scenarios.
This paper proposes Graph-as-Memory Tuning (GMT), a novel framework that enhances Knowledge Graph Completion by replacing shallow prefix concatenation with a deep cross-attention mechanism that dynamically injects compressed, context-aware graph memory tokens into frozen Large Language Models for more robust reasoning.
This paper introduces OmniVideoBench, a large-scale benchmark comprising 1,000 rigorously annotated QA pairs derived from 628 diverse videos, designed to comprehensively evaluate the synergistic audio-visual reasoning capabilities of multimodal large language models and highlight the significant performance gap between current models and human reasoning.
This paper introduces XFactor, the first geometry-free self-supervised model that achieves true novel view synthesis by disentangling camera pose from scene content through a novel augmentation scheme, thereby enabling pose transferability across different 3D scenes without relying on explicit 3D inductive biases.
This paper demonstrates that narrow finetuning leaves distinct, interpretable biases in LLM activations that can be extracted via model diffing to reconstruct training data characteristics and enhance interpretability, while warning that such models may not accurately represent broader finetuning scenarios and suggesting that mixing pretraining data can mitigate these overfitting traces.
This paper introduces CBF-RL, a framework that integrates Control Barrier Functions directly into the reinforcement learning training process to internalize safety constraints, thereby enabling safe, robust, and filter-free deployment of policies in real-world scenarios like humanoid robot navigation.
The paper introduces MSSR, a dual-agent framework that enhances Vision-Language Models' spatial reasoning by programmatically curating a Minimal Sufficient Set of 3D perception results from expert models to overcome bottlenecks caused by inadequate 3D understanding and redundant information, thereby achieving state-of-the-art performance on challenging benchmarks.
This paper introduces SceneCOT, a novel framework that achieves grounded question-answering in 3D scenes by decoupling complex reasoning into manageable steps with visual clues, supported by the newly created SCENECOT-185K dataset, which demonstrates state-of-the-art performance and represents the first successful application of Chain-of-Thought reasoning to 3D scene understanding.
The paper introduces Schrödinger Bridge Mamba (SBM), a novel one-step speech enhancement model that synergizes the Schrödinger Bridge training paradigm with the Mamba architecture to achieve superior denoising and dereverberation performance with real-time feasibility, outperforming strong generative and discriminative baselines.
The paper introduces Grasp Any Region (GAR), a multimodal large language model that leverages RoI-aligned feature replay to integrate global context for precise, interactive region-level understanding and compositional reasoning, achieving state-of-the-art performance on both image and video benchmarks.