News

Researchers introduced PolyInterview, an LLM-based platform that provides immersive mock interview practice with multimodal assessment. The system generates tailored questions from job descriptions and CVs, conducts spok…

Researchers applied dynamical systems analysis to latent chain-of-thought reasoning methods CODI and COCONUT, revealing that CODI behaves as a stable attractor while COCONUT exhibits unstable expanding dynamics. The stud…

Researchers introduced a hierarchical human-AI triage model for fraud detection in Nigeria's FinTech sector that reduced regional performance gaps from 19.43 to 2.88 percentage points, neutralizing structural bias agains…

A benchmark study found that YOLO26, despite its NMS-free end-to-end inference and CPU optimization, offers no significant accuracy advantage over predecessors YOLOv5u, YOLOv8, and YOLO11 for fish mortality detection in …

Researchers propose a dynamic scene graph attention framework (DSiGAT) that predicts lane-change intentions and trajectories for multiple interacting vehicles in traffic scenes. Tested on NGSIM and highD datasets, DSiGAT…

Researchers introduced BatteryLake, a data lakehouse that uses LLM agents and physics-grounded rules to curate heterogeneous battery aging datasets into benchmark-ready assets, releasing a benchmark of 41 datasets from o…

Researchers propose RSLoRA, a training-free rank allocation method for LoRA that uses activation-space geometry to assign ranks based on representational sensitivity. The method outperforms existing allocators like AdaLo…

Researchers introduced DUNE, a training-free refinement framework for diffusion models that detects and suppresses abrupt fluctuations in deep internal latents, improving fidelity and reducing hallucinations across U-Net…

Researchers proposed a knowledge-constrained shape-optimization framework using a Mixture-of-Experts Neural Operator (MoE-NO) for aerodynamic design, achieving up to 10% drag reduction in vehicle shape optimization. The …

Researchers introduced WiCAT, a multi-subject model for widefield calcium imaging that uses self-supervised pretraining and atlas-grounded tokenization to outperform single-session models. The model enables zero-shot beh…

Researchers proposed a generalized deep non-negative matrix factorization (G-DNMF) method for synthetic aperture radar (SAR) automatic target recognition that discards layer-by-layer decomposition to avoid error accumula…

Researchers identified two mechanisms behind effective stored-noise diffusion inversion: element-wise compression asymmetry, where int8 full-dimensional anchors preserve reconstruction but low-dimensional subspace summar…

Researchers introduced the first diffusion-based unrestricted adversarial attack against LiDAR range-image segmentation, using adversarial guidance from a segmentation loss to generate realistic yet misleading examples. …

Researchers introduced a memory-conditioned tool-calling system for camera-first visual agents that uses a three-layer personal visual memory to improve tool selection and argument generation. In tests on 800 images, rem…

Researchers introduced Prompting-MammAlps, the first camera-trap text-to-video retrieval benchmark, and a fine-grained method that uses a vision transformer for spatiotemporal action localization and an LLM-based coding …

Researchers introduced AuditWeave, a lightweight Python library that records AI-assisted and data-transformation workflows into an append-only, hash-chained ledger for tamper-evident auditing. The tool enables reviewers …

A comprehensive survey proposes a two-level taxonomy for graph neural network (GNN) methods across the knowledge graph pipeline, covering construction, embedding, reasoning, and applications. The review categorizes appro…

A new position paper argues that ground truth datasets in machine learning are human constructions, not objective truths, and calls for the ML community to acknowledge their contingent, context-dependent nature to improv…

Researchers introduced MawForge, a system for running sparse Mixture-of-Experts language models on memory-constrained local machines by storing the full model on disk and materializing expert tensors into a bounded cache…

Researchers introduced a reference-based method to detect whether a large language model was distilled from a stronger teacher model, achieving near-perfect accuracy in identifying the teacher in single-teacher scenarios…