Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers introduced a visual analytics framework for exploring attention dynamics in diffusion models, enabling structured analysis of token-level cross-attention maps across generation steps. The framework integrates…
Researchers propose a reinforcement learning with verifiable rewards (RLVR) method for data-efficient adaptation of audio-language models to code-switched automatic speech recognition (ASR). Using Qwen2-Audio across 10 l…
Researchers at a university found that selectively applying monocular depth priors from Depth Anything V2 improves sparse-view neural reconstruction in driving scenes, with Splatfacto gaining 1.029 PSNR and reducing RMSE…
Researchers used interpretable machine learning to predict Parkinson's disease motor severity from QSM and fMRI features, achieving 45.4% variance explained and 75% of predictions within 5 points of clinical scores. The …
Researchers introduced VulcanVoxel, a spatial affordance model that learns 3D blade insertion feasibility in cluttered stowing environments using a masked autoencoder over occupancy fields. Trained on 10,000 real warehou…
Researchers developed a two-stage U-Net framework for efficient pedestrian-level wind prediction across urban morphologies, using the UrbanTALES dataset. The first stage predicts wind fields patch-by-patch, while the sec…
Researchers introduced SkillOpt-Lite, a minimal skill optimization pipeline for autonomous agents that accelerates convergence and improves performance, achieving +8.8 points on LiveMath with GPT-5.5 and +25.4 points wit…
Researchers evaluated uncertainty-aware last-layer adaptation methods for referable diabetic retinopathy screening using RETFound, a self-supervised vision-transformer foundation model, on the APTOS 2019 and DDR datasets…
A new decision framework for cross-habitat marine species recognition shows that pairing a frozen DINOv2 foundation model with a linear classifier and just 10-20 labeled images per species achieves reliable recognition a…
A cross-dataset benchmark of CNN and Transformer architectures for diabetic foot ulcer segmentation found that all models degrade substantially on external data, but SegFormer-B2 (Transformer) generalizes best, outperfor…
Researchers propose ASK+, an uncertainty-gated framework that improves small language model (SLM) guidance for reinforcement learning agents under partial observability. By providing trajectory-aware context and structur…
IFLYTEK-Embodied-Omni, a unified multimodal foundation model, jointly models vision, language, and action within a single framework to enable general-purpose embodied agents to understand instructions, predict environmen…
Researchers proposed DMKGC, a conditional diffusion-guided knowledge transfer framework for multi-domain knowledge graph completion, achieving a 4.3% average MRR improvement over state-of-the-art methods. The framework t…
Researchers propose TR-RAG, a teacher-regularized reinforcement learning method for cross-lingual retrieval-augmented generation that addresses language drift and unreliable evidence use. The method combines reward optim…
Researchers developed a multilingual, multimodal NLP framework for early detection of fake news and violence-driven mob activity, achieving 98% accuracy on a fused dataset of 138,256 Bangla and English samples. The syste…
Researchers propose MAGE, a geometry-aware framework for view-guided point cloud completion that integrates efficient modality alignment and adaptive geometry enhancement to address cross-modal geometric inconsistency. T…
Researchers propose a fuzzy-logic extension of Answer Set Programming (ASP) to bridge numerical data and qualitative reasoning, using learning-based membership functions and semantically enriched predicates. The framewor…
Researchers propose EmCom-Diffusion, a framework that measures visual reflection in emergent languages by reconstructing images from emergent messages using a finetuned diffusion model and comparing them to originals. Va…
Researchers propose R²PO, a method that decouples rollout and inference policies for LLM reasoning, achieving accuracy gains of 3.4% on MATH-500 and 1.3% on APPS. The approach attaches a lightweight Residual Rollout-Head…
Researchers propose FlashBlock, a cached block-external attention mechanism that reduces attention computation and KV cache access in block diffusion models for long-context generation. The method achieves up to 1.44× hi…