AI News

Agaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented Languages (arxiv.org)

Resource-Efficient 3D Shape and Orientation Estimation of Road Users using Roadside RGB Camera and 3D Radar (arxiv.org)

Do Factuality Methods Make LLMs More Factual? Reproducibility and Fair Comparison (arxiv.org)

Not All Drift Is a Mechanism: Tracing Prompt Adaptation in LLMs (arxiv.org)

Graph-free Retrieval with Graph-Modified Embeddings (arxiv.org)

PRISM: Calibrating Black-Box Judges with Counterfactual Probes (arxiv.org)

Urbex: Agentic Spatial Grounding for City-Scale 3D Scenes (arxiv.org)

Let the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic Manipulation (arxiv.org)

Lightweight but Linguistically Fair? Quantisation Robustness across Bavarian, German, and English (arxiv.org)

ShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-On (arxiv.org)

SimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender Systems (arxiv.org)

AffectWorld: Benchmarking Emotion-Driven Future World Prediction (arxiv.org)

Teaching MLP More Graph Information: A Three-stage Multitask Knowledge Distillation Framework (arxiv.org)

Which Optimizer Leaks Most? Optimizer Implicit Bias as a Training-Data Privacy Controller (arxiv.org)

JevGPT: Thinking Fast and Slow in Autonomous Research (arxiv.org)

Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher (arxiv.org)

A Few More Tokens Ahead: Reinforcing Parallelism in Diffusion LLMs (arxiv.org)

Looped Reasoning Finishes Earlier Than You Think (arxiv.org)

LoOPSD: On-Policy Self-Distillation with Looped Teachers (arxiv.org)

Claim Framing as a Driver of LLM Sycophancy (arxiv.org)

When Do Learned Blocks Preserve Causal Meaning? Block Identifiability and Constructive Causal Abstractions (arxiv.org)

H-DDCM: Hadamard Codebooks for Fast Generative Image and Video Compression (arxiv.org)

Certifying Almost-Sure Reachability in Unknown MDPs (arxiv.org)

When Missingness Follows the Graph: Beyond Feature Propagation for Missing Features (arxiv.org)

A CaLLM Judge is a Better Judge: Mitigating Bias in LLM-as-a-judge (arxiv.org)

APT: Feedback-Driven Agentic Post-Training for Text-to-Image Generation (arxiv.org)

Uncertainty Evaluation: Bridging Machine Learning and Measurement Science (arxiv.org)

Lifting Finite-Sample Gaussian Approximation of Wide Neural Networks to Functional Convergence (arxiv.org)

From Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-Learning (arxiv.org)

Fan Out or Relay? Ownership-Aware Routing for LLM Agents Querying Siloed Tables (arxiv.org)

inLoRA-MINT: Internal Representations for Auditing Training Data in Domain-adapted LLMs (arxiv.org)

High-fidelity Three-Dimensional Flow Field Reconstruction Driven by Turbulent Fluid Dynamics (arxiv.org)

Can Machine Learning Potentials Be More Accurate Than the Methods They Are Trained On? (arxiv.org)

ChampionKV: Head-Wise Adaptive Budget Allocation for KV Cache Compression (arxiv.org)

LatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent Systems (arxiv.org)

Correcting or Continuing? What On-Policy Distillation Learns from Student Errors (arxiv.org)

Cost-Efficient Privacy Enforcement for Email-Generating LLM Agents under Strategic Information Requests (arxiv.org)

Match the Canvas, Not Just the Answer: Predictability-Aware SFT for Diffusion Language Models (arxiv.org)

Saliency-Guided Replay for Continual Learning in Deep Spiking Neural Networks (arxiv.org)

Seeing Through the Forgery Is Not Enough: Regressive Sycophancy under Fabricated Evidence in Medical Vision-Language Models (arxiv.org)

BELT : BEhavior-Aligned Low-Rank KV Cache Compression with Learned Top-k Rank Allocation (arxiv.org)

LLM Agents for Automated Post-Training via Structured Recipe Optimization (arxiv.org)

Iterative Critique-Refine Framework for Enhancing LLM Personalization (arxiv.org)

Rethinking Langevin Thompson Sampling from A Stochastic Approximation Perspective (arxiv.org)

Cognifying Education: Mapping AI's transformative role in emotional, creative, and collaborative learning (arxiv.org)

PPGFlowECG: Latent Rectified Flow with Cross-Modal Encoding for PPG-Guided ECG Generation and Cardiovascular Disease Detection (arxiv.org)

SciRerankBench: Benchmarking Rerankers Towards Scientific Retrieval-Augmented Generation with LLMs (arxiv.org)

Benchmarking and Training Reward Models for Cross-Source Evidence Grounding in Search Agents (arxiv.org)

Empowering Edge Agents: A Systematic Analysis of Edge-Cloud Collaboration in LLM Agent Execution (arxiv.org)

On Making $\mathrm{SE}(2)$-Invariant Networks Optimal (arxiv.org)