Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
SinAE, a new AI model from an unnamed research team, unifies molecules, crystals, and proteins into a single Transformer-based architecture, achieving near-lossless reconstructions via flow-matching decoding. By reducing…
A new approach using sparse autoencoders (SAEs) is revolutionizing out-of-distribution (OOD) detection by extracting interpretable features from intermediate neural network layers, achieving record performance on OOD det…
A new training-free attribution method reveals that Feedforward Networks in Transformer models exhibit sparse, structured inter-layer dependencies, where only a small subset of preceding neuron activations significantly …
ConDitar-dev, an advanced drug design framework, achieved a remarkable average binding score of -8.85 kcal/mol and enhanced ADMET properties by 73%, outperforming state-of-the-art SBDD methods. In real-world tests on PD-…
New research reveals that up to 77% of features in degraded Sparse Autoencoders (SAEs) are functionally inert, and even in well-trained models 9% of matched features are dead, challenging the interpretability claims of S…
A new AI-driven approach to solving partial differential equations learns where to add mesh resolution before field evolution, outperforming fixed-grid methods across five PDE scenarios. The two-stage diffusion framework…
A new machine learning model called NAVIS, developed using temporal graph techniques on SEC Form 13F filings, predicts institutional portfolio allocations with a test NDCG of 0.9127, outperforming all other dynamic graph…
A new AI-driven beamforming framework using an enhanced in-context learning Transformer backbone with encoder-decoder networks outperforms existing methods in multi-user MISO systems without retraining across different c…
Graph-constrained AI models outperform traditional methods in clinical code prediction, achieving a 0.709 micro-F1 score on a curated 50-code subset and 0.527 on the full 15,761-code space of the MIMIC-IV dataset, with i…
A new semi-supervised learning framework introduces unbiased risk estimators built from linear combinations of component risks, extending beyond binary classification into multiclass classification. The framework claims …
A recent analysis of Chronic Kidney Disease prediction studies found that high-leakage studies reported accuracy rates up to 95.48%, while studies without data leakage achieved around 80.2%, revealing a 15% gap due to me…
Researchers have developed BattVAE-GP, a hybrid framework that blends physics-based simulations with probabilistic learning to predict lithium-ion battery degradation more efficiently. The model uses a DFN/P2D electroche…
Support Vector Attention (SV-Attention), a new approach to memory in machine learning, achieves certified token deletion with a rare-item recall of 0.86 compared to 0.32 for other methods, according to researchers. The t…
Researchers at arXiv improved the Huth et al. fMRI encoding baseline by expanding voxel selection to 15K and substituting GPT-2 medium for GPT-1, achieving mean METEOR = 0.149 and BLEU-1 = 0.200 for subject UTS03, an 11%…
A new anatomy-privileged teacher-student framework for predicting perineural invasion from T2-weighted MRI achieved the highest mean AUROC of 0.750 among matched MRI-only baselines in a study of 155 patients, according t…
Researchers propose SymbOmni, an agentic omni-model that uses Symbolic Concept Learning to overcome the 'perpetual novice' problem in visual generation, enabling cumulative evolution and autonomous self-improvement. The …
Researchers propose Self-Consistent Flow (SC-Flow), a new method that unifies velocity and endpoint prediction in rectified-flow generative models, achieving improved generation quality and training stability. The method…
Researchers introduced Semidirect Fourier Delta Attention (SFDA), a phase-controlled generalization of Kimi Delta Attention that replaces real diagonal decay with block-rotational Fourier control, achieving exact affine …
A study comparing the RG-Flow Transformer, which incorporates a renormalization-group inductive bias, against a parameter-matched vanilla transformer on scarce EEG data from the PhysioNet Sleep-EDF corpus found no signif…
A new graph-based framework from arXiv research (2607.11894v1) detects disinformation narratives in Telegram ecosystems by combining weak supervision with propagation graph analysis, enabling identification of coordinate…