All entries

What are Key Factors for Updates in RL for LLM Reasoning?

arXiv:2606.22570v2 Announce Type: replace Abstract: Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as a promising framework for enhancing the reasoning ability of large language mod

safetyarxiv-cs-cl

What Can a Recurrent State Safely Forget?

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

arXiv:2609.23366v1 Announce Type: new Abstract: Recurrent models must preserve information that changes future behavior while suppressing hidden-state error. These objectives conflict: contraction imp

safetyarxiv-cs-lg

22 Sept 2026

What do VLM-Based Vision-Language Navigation Models Rely on: Interpreting and Steering Policy Behavior

arXiv:2609.24576v1 Announce Type: cross Abstract: Modern Vision-Language Navigation (VLN) models rely mostly on pre-trained large Vision-Language Models (VLMs) to predict navigation actions. While thi

safetyarxiv-cs-cv

What Is The Political Content in LLMs' Pre- and Post-Training Data?

arXiv:2509.22367v3 Announce Type: replace Abstract: Large language models (LLMs) reflect politically-slanted opinions in their generated text. Even though it is widely assumed that model behavior stem

What Makes a Good Medical Image Tokenizer? Rethinking Reconstruction and Generation in Medical Image Tokenization

arXiv:2609.24691v1 Announce Type: new Abstract: Latent diffusion models now dominate medical image generation, and every such pipeline rests on a tokenizer that compresses images into the latent codes

What Makes Good Multilingual Reasoning? Disentangling Traces with Measurable Features

arXiv:2604.04720v2 Announce Type: replace Abstract: Large Reasoning Models (LRMs) still exhibit large performance gaps between English and other languages, yet much current work assumes these gaps can

Tutorials

What Matters in Designing World Action Models: An Empirical Study

arXiv:2609.24048v1 Announce Type: new Abstract: World Action Models (WAMs) have emerged as a promising paradigm for generalizable robot control. Despite the growing number of WAM systems, existing wor

tutorialsarxiv-cs-ro

When and Why Do Linear Bias Probes Fail? A Geometric and Statistical Theory of Bias Detectability in Large Language Model Representations

arXiv:2609.22337v1 Announce Type: cross Abstract: Linear probing is the standard instrument for detecting social biases in the hidden representations of large language models. Yet reported probe accur

safetyarxiv-cs-lg

When Cosine Similarity Fails to Reflect Linearly Accessible Structure in Dialogue Models

arXiv:2609.22522v1 Announce Type: new Abstract: Cosine similarity is widely used to analyze transformer representations, implicitly assuming that similarity reflects task-relevant structure. We study

Agents

When Digital Twins Meet Large Language Models: Realistic, Interactive, and Editable Simulation for Autonomous Driving

arXiv:2507.00319v3 Announce Type: replace Abstract: Simulation frameworks have been key enablers for the development and validation of autonomous driving systems. However, existing methods struggle to

agentsarxiv-cs-ro

When Does Adversarial Refinement Help? A Negative Result and Open Problem in Adapting R3GAN to Time Series Imputation

arXiv:2609.23102v1 Announce Type: new Abstract: Diffusion models and transformers have supplanted GANs for multivariate time series imputation, largely on grounds of GAN training instability. R3GAN (N

When Does Learning Beat Heuristics? A Case Study in Kubernetes Scheduler Score Plugins

model-releasesarxiv-cs-lg

arXiv:2609.22142v1 Announce Type: cross Abstract: Kubernetes scheduler plugins that score candidate nodes are, in production, hand-tuned heuristics. We ask whether a learned scoring function - trained

When Does Test-Time Physical Diagnosis Pay? A Frozen Policy Buys Evidence It Never Reads

arXiv:2609.22299v1 Announce Type: cross Abstract: When a robot faces unfamiliar physical conditions, a common approach is to collect evidence about what changed and adapt. For such diagnosis to improv

safetyarxiv-cs-lg

When Does Touch Matter? Charting the Vision-Interaction Gap in Cluttered Dexterous Grasping

model-releasesarxiv-cs-ro

arXiv:2609.24068v1 Announce Type: new Abstract: Dexterous grasping in clutter poses a basic sensing question: when do tactile measurements and external wrench estimates improve on visual geometry? Occ

When Evidence Conflicts: Reliability-aware Meta-review Generation

arXiv:2609.24028v1 Announce Type: new Abstract: Generating coherent meta-reviews from multiple peer reviews is challenging when reviewer evidence conflicts and varies in reliability. Existing approach

When is a closed-form RGB->S/P ratio adequate? A hyperspectral characterization on natural scenes for mesopic display

arXiv:2609.24819v1 Announce Type: cross Abstract: Mesopic and low-light display transforms require, as their driving signal, a per-pixel scotopic-to-photopic luminance ratio (S/P); the exact spectral

When Is Availability-Aware Training Worth It? A Benchmark and Empirical Study of Interruption-Resilient Optimization Under Predictable Compute Schedules

model-releasesarxiv-cs-lg

arXiv:2609.22087v1 Announce Type: cross Abstract: Training under non-stationary but predictable compute availability (satellites under eclipse, duty-cycled edge devices, power-capped datacenters) is o

When Quantization Preserves Accuracy but Not Evidence: Explanation-Aware Post-Training Quantization for Medical LLMs

arXiv:2609.24799v1 Announce Type: new Abstract: Post-training quantization (PTQ) enables efficient deployment of large language models, and PTQ methods are usually optimized and evaluated with generic

When Residualization Helps an Audit: Format Effects, Slice Gains, and Their Limits

arXiv:2609.24194v1 Announce Type: new Abstract: Evaluation scores used around LLM systems -- including reward models, rerankers, and LLM judges -- can track surface form instead of the quality they cl

safetyarxiv-cs-cl

When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series Forecasting

arXiv:2609.24862v1 Announce Type: new Abstract: Agentic time series forecasting concerns systems whose underlying mechanisms evolve, making the relative effectiveness of numerical models, reasoning st

safetyarxiv-cs-lg

When Who You Are Can Change the Code You Get: A Study of Persona-Induced Bias in LLM Code Generation

arXiv:2609.22102v1 Announce Type: cross Abstract: Large Language Models (LLMs) are widely used as programming assistants, yet it remains unclear whether and how user's demographic information impacts

When Wider Views Fail: Stress-Testing Feed-Forward 3D Reconstruction

arXiv:2609.24839v1 Announce Type: new Abstract: Feed-forward 3D reconstruction models enable efficient geometry estimation from sparse images, but their pretrained nature can make them vulnerable to d

Which Part of the Context Layer Does the Work? Separating Semantic Content from Retrieval Scaffolding in Text-to-SQL Agents

arXiv:2609.22259v1 Announce Type: new Abstract: Context layers, curated documentation that an analytics agent fetches at query time, produce large accuracy gains on text-to-SQL benchmarks. A with/with

Which Terrain Is Better? Preference Learning with VLM Prototypes for Off-Road Traversability Ranking

arXiv:2609.23673v1 Announce Type: new Abstract: In vision-based off-road navigation, a robot needs to know not only which obstacles to avoid but also which terrain is better. The first is handled by f

Whitening Inverts the Hierarchy: What the Norm of a Whitened Embedding Measures

arXiv:2609.23117v1 Announce Type: new Abstract: Whitening a foundation-model embedding and using its squared norm as a training-free likelihood surrogate is motivated by the observation that whitened

Whole-Body UMI: Transferring UMI Manipulation Skills to Humanoid Whole-Body Manipulation via Real-Time Motion Generation

arXiv:2609.22829v1 Announce Type: new Abstract: Collecting whole-body demonstrations for humanoid manipulation mostly relies on teleoperation, which is costly and hard to scale up. The Universal Manip

safetyarxiv-cs-ro

Applications

Why Do Video Diffusion Models Violate Physics? Unveiling the Flaws in Attention Mechanisms

arXiv:2609.23658v1 Announce Type: new Abstract: Despite impressive visual quality, state-of-the-art video diffusion models often generate content that violates real-world physical laws. While existing

applicationsarxiv-cs-cv

Why Ghost Outputs Teach: A Kernel-Based Understanding of Subliminal Learning

arXiv:2609.23260v1 Announce Type: new Abstract: Subliminal Learning (SL) is a recently identified phenomenon in which a student model acquires downstream task capabilities by matching seemingly unrela

WildfireSpreadBench: The Metric Decides the Model in Wildfire Spread Prediction

arXiv:2609.22191v1 Announce Type: cross Abstract: Machine learning is being increasingly used to predict where active wildfires will burn the following day, helping inform evacuation boundaries and co

Applications

WiNeRF: Measurement Constrained Radiance Fields for Actionable Wireless Channel Modeling

arXiv:2609.22132v1 Announce Type: cross Abstract: Wireless embedded systems increasingly rely on wireless channel information for decision making, yet practical platforms operate under severe constrai

applicationsarxiv-cs-lg

Local Ai

WOLF: World Model Guided LiDAR Exploration with Predictive Frontiers

arXiv:2609.23656v1 Announce Type: new Abstract: LiDAR-based unmanned aerial vehicle (UAV) exploration builds maps by continually selecting where to observe next. However, decisions based on the measur

local-aiarxiv-cs-ro

WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory

arXiv:2609.24984v1 Announce Type: new Abstract: Video world models enable interactive exploration of dynamic environments, yet struggle to respect prior observations over long horizons and across view

WorldScape Policy 2.0: Empowering Steerable World Action Modeling with Reasoning-Augmented Memory and In-Context Learning

arXiv:2607.18840v2 Announce Type: replace Abstract: World Action Models (WAMs) offer a promising paradigm for robotic manipulation by jointly modeling visual state transitions and robot actions. Howev

safetyarxiv-cs-ro

WPBench: A Comprehensive Benchmark for Wind Power Forecasting

model-releasesarxiv-cs-lg

arXiv:2609.24444v1 Announce Type: new Abstract: Accurate, reliable, and deployable wind power forecasting is critical for power system dispatch, renewable energy integration, and electricity market op

Written as a Record, Read as an Address: What a Forward Pass Leaves in an Operation's KV Cache

arXiv:2609.24635v1 Announce Type: new Abstract: When a language model reads an operation such as 'Swap the contents of Box F and Box B', its forward pass writes keys and values for those tokens into t

X-Beat: An Explainable Framework for ECG Image Classification

arXiv:2609.22631v1 Announce Type: new Abstract: Accurate automated interpretation of electrocardio- grams (ECGs) is essential for early detection of cardiac condi- tions such as myocardial infarction

Local Ai

XSQ-AST: An Explainable Audio Spectrogram Transformer Framework for Localising Synthetic Speech Artifacts

arXiv:2609.24770v1 Announce Type: cross Abstract: Localising artifacts in synthetic speech remains challenging, as most evaluation methods yield only global quality scores. This paper presents XSQ-AST

local-aiarxiv-cs-lg

XYEval: Agents say yes to bad advice

arXiv:2609.23939v1 Announce Type: new Abstract: Effective communication between users and AI agents is essential for human-AI collaboration. The XY problem is a well-known communication pitfall where

Yarn tracking of large-scale 3D textile reinforcements using topological material features

arXiv:2609.22315v1 Announce Type: new Abstract: Automated segmentation of CT images has become increasingly important to enhance the reliability of simulations through the generation of high fidelity

You Can Tell Who's Asking: What the Web's Questions Are Made Of, and Where They Come From

arXiv:2609.24106v1 Announce Type: new Abstract: Questions scraped from the web are used across academia and industry as a proxy for what people want to know. Across QA training data, retrieval benchma

Zeva-Ego: Egocentric Mid-Training with In-Context Causal Learning for Robot Manipulation

model-releasesarxiv-cs-ro

arXiv:2609.24411v1 Announce Type: new Abstract: Egocentric video offers a scalable source of physical interaction experience, yet translating it into robot-executable knowledge and enabling continual

Agents

ZIL: Zero-shot Image-to-LiDAR Registration

arXiv:2609.22716v1 Announce Type: new Abstract: Image-to-LiDAR registration estimates the camera pose of an image with respect to a LiDAR point cloud. It has diverse applications in autonomous driving

agentsarxiv-cs-cv

ZoAQ: Adaptive Zeroth-Order Querying via Query-Reuse Coupling

arXiv:2609.22115v1 Announce Type: new Abstract: Zeroth-order optimization (ZOO) estimates updates from function evaluations, making perturbation queries a primary cost. Fixed budgets spend the same nu

ZVeC: A Zero-Shot Framework for Instance-Level Vehicle Extraction and Generative Point Cloud Completion

arXiv:2609.24825v1 Announce Type: new Abstract: LiDAR point clouds acquired in underground environments exhibit severe geometric incompleteness due to occlusions and limited sensor viewpoints, making

2D GauSS-MI: Efficient Active Scene Reconstruction with Balanced Visual and Geometric Quality

arXiv:2609.21516v1 Announce Type: new Abstract: Active reconstruction requires efficient active view selection to achieve high-quality reconstruction within limited onboard computational resources. Ex

2nd Place Solution to the HANDS 2026 Workshop Challenge-Dexterous Grasp Motion Track: Single-Shot Trajectory Warping for Grasp Motion Generation

arXiv:2609.21511v1 Announce Type: cross Abstract: This report describes our 2nd place solution to the HANDS 2026 workshop challenge (Dexterous Grasp Motion track) in conjunction with ECCV 2026. In thi

safetyarxiv-cs-ai

4DGS-Fixer: Generative Sparse-View 4D Gaussian Splatting with Iterative Refinement Guided by Video Diffusion Priors

arXiv:2609.21176v1 Announce Type: new Abstract: This paper addresses the challenges of dynamic scene synthesis from sparse-view videos. Existing methods employ geometric priors, adaptive optimization,

A benchmark dataset and baseline methods for four-dimensional STEM diffraction patterns

arXiv:2609.21593v1 Announce Type: new Abstract: Four-dimensional scanning transmission electron microscopy (4D-STEM) records a two-dimensional diffraction pattern at each electron-probe position, yiel