Search: “research”

The Optimal Decoder is Tilted Temperature

arXiv:2610.15058v1 Announce Type: new Abstract: Larger sample budgets benefit from higher sampling temperatures in LLMs, but what temperature tuning optimizes and why it works remain poorly understood

The Surprising Expressivity of a Single Neuron with Autoregressive Chain-of-Thought

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

DGX agent

arXiv:2610.17283v1 Announce Type: new Abstract: The success of autoregressive language models trained with Chain-of-Thought~(CoT) supervision raises a fundamental question: How simple can a next-token

researcharxiv-cs-lg

14 Oct 2026

The Value of Being Wrong: Self-Mined Visual In-Context Learning from Errors

arXiv:2610.16977v1 Announce Type: new Abstract: Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depends c

Thinking with Images without Coordinates

arXiv:2610.14265v1 Announce Type: new Abstract: Recent advances in multi-modal large language models have shifted toward the 'thinking with images' paradigm to improve reasoning faithfulness, yet exis

TIDE: Trajectory-Informed Skill Distillation for Efficient Text-to-Video Prompt Optimization

arXiv:2610.15098v1 Announce Type: new Abstract: Text-to-video (T2V) models are highly sensitive to prompt specificity, while iterative prompt optimization typically improves generation quality only th

Tight End-to-End Accounting for Differentially Private Synthetic Data

arXiv:2610.17249v1 Announce Type: cross Abstract: Differentially private (DP) pipelines often combine heterogeneous mechanisms, yet their guarantees are best analyzed using different notions of privac

ToF-Wheel Odometry: Navigating Degeneration of Localization With Limited Sensing Capabilities

arXiv:2610.14590v1 Announce Type: cross Abstract: Wheel odometry provides an efficient motion prior for ground robots but is prone to transient violations such as wheel slip or terrain-induced tilt. I

Token-Budgeted Candidate Retrieval for 3D Scene-Grounded VLMs: Category Names vs. Pairwise Geometry

arXiv:2610.14207v1 Announce Type: new Abstract: Under a fixed context budget, which objects and relations should a 3D scene-grounded VLM see first? On ScanNet indoor referring benchmarks (Dai et al.,

Totem: Object-Centric 3D Memory for World-Consistent Video Generation

arXiv:2610.15807v1 Announce Type: new Abstract: Generating coherent videos grounded in a unified underlying world remains a major challenge, particularly under large viewpoint changes, where existing

Toward Guarantees for Clinical Reasoning in Vision Language Models via Formal Verification

arXiv:2602.24111v2 Announce Type: replace Abstract: Vision-language models (VLMs) show promise in drafting radiology reports, yet they frequently suffer from logical inconsistencies, generating diagno

Toward Study on the Effects of Robot's Humorous Apology on Human's Trust Recovery

arXiv:2610.16334v1 Announce Type: new Abstract: Robot errors can disrupt human-robot collaboration by reducing user trust and willingness to rely on the robot. Robot apologies can support trust recove

researcharxiv-cs-ro

Towards Complementary Keypoint Detection via Mixture of Detectors

arXiv:2610.16762v1 Announce Type: new Abstract: Keypoint detection plays a central role in geometric vision pipelines. Existing detectors exhibit complementary strengths across different scene structu

TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generation

arXiv:2610.15365v1 Announce Type: cross Abstract: Large language models (LLMs) can produce harmful content while generating a response. A useful safeguard should not only detect this risk during gener

TraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative Sampling

arXiv:2610.14332v1 Announce Type: new Abstract: Accurate sampling from continuous-time generative models can require many model evaluations. Low-step sampling reduces cost, but coarse updates introduc

Train for Many, Update with One: Improving Multi-Sample Reasoning via Fenchel-Bregman Posterior Learning

arXiv:2610.17735v1 Announce Type: new Abstract: Multi-sample reasoning measures candidate coverage, the probability that at least one of K candidates is correct. However, minimizing the average loss o

Training-Free Dense Hand Contact Estimation with Multi-Modal Large Language Models

arXiv:2605.05886v2 Announce Type: replace Abstract: Dense hand contact estimation requires both high-level semantic understanding and fine-grained geometric reasoning of human interaction to accuratel

TrajCFG: Token-Wise Classifier-Free Guidance via Denoising Trajectory Feedback

arXiv:2610.14153v1 Announce Type: new Abstract: Classifier-free guidance (CFG) typically applies a shared scale to all latent tokens, although their velocity predictions evolve differently during samp

TropNNC: Structured Neural Network Compression Using Tropical Geometry

arXiv:2409.03945v4 Announce Type: replace Abstract: We present TropNNC, a framework for compressing neural networks with linear and convolutional layers and ReLU-type activations using tropical geomet

TwinPrune: Density-Aware Two-Phase Token Pruning for Vision-Language Models

arXiv:2610.17530v1 Announce Type: new Abstract: We propose a unified spatial-density signal to score token importance for both inference phases: bipartite merge for prefill and KV eviction for decode,

Two Linear Programs are Better than One

arXiv:2610.15909v1 Announce Type: cross Abstract: The extension complexity operatorname{xc}(P) of a polytope P is the minimum number of facets of a polytope that projects onto it. Motivated by the com

ULPack: Reference-Conditioned Lossless Compression of Checkpoint Trajectories

arXiv:2610.14184v1 Announce Type: new Abstract: Studying model training requires checkpoint trajectories, whose storage cost limits temporal resolution. Although adjacent checkpoints are numerically c

UMAS3D: Unified Mask-Free 3D Editing with Region-Adaptive Source Preservation

arXiv:2610.15533v1 Announce Type: new Abstract: 3D editing aims to modify an existing 3D asset according to user intent while preserving geometry and appearance outside the intended edit region. Exist

Uncovering the Generative Landscape of Diffusion Models: A Spectral-Guided Quality-Diversity Diffusion Framework

arXiv:2610.15724v1 Announce Type: new Abstract: Denoising Diffusion Probabilistic Models have emerged as powerful generative priors across domains ranging from molecular design to high-resolution imag

Understanding and Mitigating Training Instability in Large-Scale Recommender Systems: A Cascade Perspective

arXiv:2610.14394v1 Announce Type: cross Abstract: Training large-scale recommender systems on dynamic industrial platforms, such as Short-Form Video (SFV) services with billions of daily interactions,

Understanding Gene Expression-to-Image Learning: Spatial Structure and Information Fidelity

arXiv:2610.17699v1 Announce Type: cross Abstract: Single-cell RNA sequencing (scRNA-seq) enables cell-level analysis of biological heterogeneity, but learning effective representations from high-dimen

Understanding Personalization Gains in Whisper

arXiv:2610.17019v1 Announce Type: cross Abstract: Whisper-style ASR models provide strong zero-shot transcription, but their residual errors are not evenly distributed across datasets, recording condi

Understanding the Interplay between Memorization and Learning in Large Language Models

arXiv:2507.14777v2 Announce Type: replace Abstract: We investigate foundational questions about the interplay between memorization and learning in large language models (LLMs). When an LLM generates a

UNISON: A Unified Framework for Learning from Scored Bags

arXiv:2610.15564v1 Announce Type: new Abstract: We introduce UNISON, a unified framework that bridges multiple instance learning and recommendation systems by treating both as instances of learning fr

Universal set of Observables for Forecasting Physical Systems through Causal Embedding

arXiv:2105.10759v4 Announce Type: replace-cross Abstract: We show how a pair of points can uniquely represent a left-infinite sequence obtained from observations of an underlying dynamical system thro

Unrolled Neural Networks for Constrained Optimization

arXiv:2601.17274v2 Announce Type: replace Abstract: We develop unrolled neural networks to solve constrained optimization problems, offering accelerated, learnable counterparts to dual ascent (DA) alg

UnSAMv2: Self-Supervised Learning Enables Segment Anything at Any Granularity

arXiv:2511.13714v2 Announce Type: replace Abstract: The Segment Anything Model (SAM) family has become a widely adopted vision foundation model, but its ability to control segmentation granularity rem

Unsupervised Data Drift Monitoring for Deployed Deepfake Speech Detectors

arXiv:2610.16826v1 Announce Type: cross Abstract: Deepfake speech detectors are typically evaluated as binary classifiers on fixed benchmarks, but deployed systems face evolving attacks, codecs, speak

Variational Filtering for Bayesian Last Layers with an Independent Binary Probit Surrogate

arXiv:2610.16578v1 Announce Type: cross Abstract: Modern deep classifiers trained by gradient descent need many passes over the data to become accurate, and even then are not always well calibrated; p

VGGA: Geometry-Grounded Autoencoder for Video Generation

arXiv:2610.16204v1 Announce Type: new Abstract: Autoencoders are the core component of latent video generative models, compressing high-dimensional videos into compact latent spaces for downstream gen

VideoJev: System One Model for Video Understanding and Interaction

arXiv:2610.18615v1 Announce Type: new Abstract: Visual agents must turn video observations into timely decisions to track progress and choose what to do next. Building on the emerging System One parad

Visually Legible but Adversarial: Watermark-Constrained Attacks on End-to-End VLMs for Document Understanding

arXiv:2610.15728v1 Announce Type: new Abstract: Visible watermarks convey document ownership, confidentiality, and status, but can also carry adversarial signals. We introduce Auto-Regional Adversaria

VLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action Models

arXiv:2610.16567v1 Announce Type: new Abstract: Vision-Language-Action models (VLAs) have achieved high success rates on familiar manipulation tasks, yet it remains unclear whether such scores reflect

VLCQ: Post-training quantization for deep neural networks using variable length coding

arXiv:2610.15092v1 Announce Type: new Abstract: Quantization plays a crucial role in efficiently deploying deep learning models on resource-constrained devices. Post-training quantization does not req

Walk Like This: Stylized Locomotion for Physics-Based Characters through Example-Driven Diffusion

arXiv:2610.18147v1 Announce Type: new Abstract: We present a real-time framework for expressive legged locomotion that enables independent control over velocity and style. Unlike existing data-driven

researcharxiv-cs-ro

Weakly Convex Ridge Regularization for 3D Non-Cartesian MRI Reconstruction

arXiv:2603.27158v3 Announce Type: replace Abstract: While highly accelerated non-Cartesian acquisition protocols significantly reduce scan time, they often entail long reconstruction delays. Deep lear

What Do Sparse Autoencoder Feature Discovery Scores Identify? From Activations to Interventions

arXiv:2610.15850v1 Announce Type: new Abstract: Sparse autoencoder (SAE) features are typically discovered from how their activations co-vary with behaviour, yet they are often used through interventi

What Limits Cross-Task Generalization in World Action Models?

arXiv:2610.16641v1 Announce Type: new Abstract: Cross-task generalization remains difficult for world action models (WAMs), even when their predicted futures depict plausible manipulation. This gap ca

researcharxiv-cs-ro

What Makes a Good Visual Representation for Spatial Understanding in MLLMs?

arXiv:2610.17844v1 Announce Type: new Abstract: Spatial understanding is a necessary step in moving multimodal large language models (MLLMs) toward reasoning about the 3D physical world. Their visual

What Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training Strategies

arXiv:2610.15035v1 Announce Type: new Abstract: Contrastive learning has become the standard paradigm for multimodal representation learning, aligning semantically matched pairs while separating misma

When Do Learned Blocks Preserve Causal Meaning? Block Identifiability and Constructive Causal Abstractions

arXiv:2610.18155v1 Announce Type: new Abstract: Standard identifiability theory assumes one-dimensional latent factors, but learned representations can encode latent information across multidimensiona

When Does Compensating Inherited Quantization Error Help? Mean-Square Guarantees for Layer-Wise Quantization and Their Limits

arXiv:2610.16020v1 Announce Type: new Abstract: Layer-wise post-training quantization methods let each layer compensate for the quantization error inherited from earlier layers. We ask under what assu

When Does Trimming Help Conformal Prediction? A Retained-Law Diagnostic under Calibration Contamination

arXiv:2605.06204v2 Announce Type: replace-cross Abstract: Trimming suspicious calibration points is a common response to contamination in conformal prediction. Its effect on clean-target coverage, how

When Dynamic Scaling Fails: A Scale-State Feedback Governs Recurrent-State Quantization in Linear-Attention and SSM LLMs

arXiv:2610.15139v1 Announce Type: new Abstract: Linear-attention and state-space LLMs carry long context in a fixed-size recurrent state, and quantizing that state is the natural way to cut its cost d