AI News

FlowBudget: Training-Free Sampling-Budget Estimation for Flow Matching Models (arxiv.org)

Spectrum Is Not Enough: A Random-Matrix Account of Data-Free Preconditioning for Differentially Private Learning (arxiv.org)

AlignSAE: Concept-Aligned Sparse Autoencoders (arxiv.org)

Feature-preserving Latent-EnKF for Data Assimilation of Flows with Shocks (arxiv.org)

Can Capability Enhancement be Localized? (arxiv.org)

DemoStream: Batched Task-and-Motion-Plan Optimization for Automated Learning from Demonstration (arxiv.org)

Priming Base Models for Agentic Post-Training: An Open Midtraining Recipe (arxiv.org)

Single-stage joint diffusion of sky images and clear-sky index for ultra-short-term irradiance forecasting: what the frame-to-irradiance stage discards (arxiv.org)

Same Spectrum, Harder First-Order Training: Phase-Locking in Single- versus Multi-Qubit Re-Uploading Circuits (arxiv.org)

Data World: Scaling Data Environments from Scratch (arxiv.org)

PocketTSLM: Deploying an On-Device Soft-Prompt Time-Series Language Model for Privacy-Preserving Smartphone Reasoning (arxiv.org)

Following Delegation Traces to Study Human Oversight in Interaction with Multimodal AI (arxiv.org)

Agentic-Priors: Using Large Language Models to Address the Cold Start Problem in Preference Elicitation (arxiv.org)

BridgeAfford: Bridging Query-Target Gap for Generalizable 3D Affordance Grounding (arxiv.org)

CodeHealth Predicts the Token Cost of Agentic Code Review: A Pre-Registered Study of 465 Agent-Authored Pull Requests (arxiv.org)

Learning Rewards in Strategic Games (arxiv.org)

Personalized Projected Ascent for Federated Stackelberg Optimization (arxiv.org)

In-Context Preference-Conditioned Pareto Set Learning from Limited Offline Data (arxiv.org)

Reducing the Cost of Qudit State Preparation by Learning a Sparse Structure for SNAP Gates (arxiv.org)

Rethinking Classifier-Free Guidance for TI2I Flow Matching (arxiv.org)

When the KV Cache Is Slower Than the GPU: Scaling Remote KV Tiers for Context-Parallel Prefill (arxiv.org)

DuET: Note-Frame Dual Encoding for Piano Performance-to-Score Transcription (arxiv.org)

Beyond Invariance: Factor-wise Selective Treatment of Shifts for Multimodal Learning (arxiv.org)

Anvil: Scaling Embodied Spatial Intelligence through Knowledge Recirculation (arxiv.org)

Cross-Embodiment Causal Transfer from Human Videos to Robot Manipulation (arxiv.org)

Decoding What iOS Location Services Send to Apple: An LLM Is an Unreliable Analyst but a Useful Toolmaker (arxiv.org)

AVALON: Zero-Shot Indoor Navigation with Human-Seeded Topological Memory (arxiv.org)

Ergodic topology sampling toward the continuum with Stochastic Normalizing Flows (arxiv.org)

What Coding Agents Leave Unsaid: Measuring Task-State Gaps Between Execution and Reports (arxiv.org)

InkSight: Offline-to-Online Handwriting Conversion by Teaching Vision-Language Models to Read and Write (arxiv.org)

LLMR: Knowledge Distillation with a Large Language Model-Induced Reward (arxiv.org)

Towards Reliable Long-Context Reasoning: Detecting Unanswerable Questions via FactGuard (arxiv.org)

AI Scientists for Building Virtual-Cell Models (arxiv.org)

LightMedSeg-ISLES: Stroke Lesion Segmentation with 81x Fewer Parameters than nnU-Net (arxiv.org)

TLA-Verus: Generating and Solving Temporal Proof Obligations with LLM Agents (arxiv.org)

PotentialEP: Backprop-Free Training of CIFAR-Scale Generative Models via Equilibrium Propagation (arxiv.org)

A self-learning framework for adaptive free-form pipe routing in aeroengines (arxiv.org)

Scientific feedback reveals productive regimes beyond training conventions (arxiv.org)

Benchmarking Code LLM-as-a-Judge via Constrained Perturbation and Calibration (arxiv.org)

Product Interaction: An Algebraic Formalism for Deep Learning Architectures (arxiv.org)

Lexical Reference Frames: Interpreting LM Residual Streams in Native Semantics (arxiv.org)

Patch where the text is new: a compression view of where byte-level models should spend compute (arxiv.org)

Beyond the Dominant Label: Recovering Coexisting Lesion Evidence with Multiple Instance Learning (arxiv.org)

The Right Byte First: How UTF-16 Helps Multilingual Boundary Prediction (arxiv.org)

LEAP: Learned Block-wise Evidence Retrieval for Long Audio-Video Perception (arxiv.org)

AVC-Reward: Event-Centric Audio-Video Consistency Evaluation (arxiv.org)

Interpretable but Fragile? Robustness of Concept Bottlenecks under Geometric-Semantic Perturbations (arxiv.org)

Fractional Power-of-Two Quantization for Efficient and Effective Multiplier-Free LLM Inference (arxiv.org)

What do AI chatbots do after users decline therapeutic redirection for suicide thoughts? (arxiv.org)

Stateful Testing for Deep Learning Compilers (arxiv.org)