Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A decade of official labor statistics from the UK Office for National Statistics and the US Bureau of Labor Statistics shows no net aggregate job destruction nearly four years into commercial large language model deploym…
NVIDIA has adopted the "Ozaki Scheme II" computational method, developed by Professor Katsuhiro Ozaki and his team at Shibaura Institute of Technology, into its CUDA toolkit 13.4, according to EE Times Japan. The scheme …
Vercel released a new version of its AI SDK for Python, adding an experimental evaluate() API that connects directly to the Jev AI model, a universal classifier that returns structured JSON answers with confidence scores…
Ortus AI released RightWayUp, an Apache-2.0 licensed 360° image rotation model that predicts how far an image is rotated (0–359°) with a confidence score and abstains when it cannot determine which way is up. The model i…
AOTInductor, PyTorch's ahead-of-time inference backend compiler, emits Triton kernels as cubin files stored in the pt2 zip archive, which developers can unzip and disassemble with nvdisasm or cuobjdump --dump-sass to ver…
A September 30, 2026 arXiv paper reports that major gains in non-invasive brain-to-text decoding are largely reproducible without any brain data, because fixed-length windows starting at each word implicitly leak word-du…
A developer built SignBridge, an open-source bidirectional real-time sign language interpreter, for a deaf friend who communicates primarily through Vietnamese Sign Language. The system runs MediaPipe hand-landmark track…
Researchers at Chalmers University of Technology, with collaborators from the University of Gothenburg and the University of Cambridge, built a closed-loop AI scientist that autonomously forms hypotheses, designs and run…
A developer built a small local decision model on a 24 GB MacBook that returns a single lettered label plus a calibrated confidence score in one forward pass, using Gemma 4 E2B (4-bit) with a 52 MB rank-16 LoRA adapter t…
A paper submitted to arXiv on 30 Sep 2026 introduces Loop Scaling Laws, which its authors describe as the first scaling law to jointly model recurrence and sparsity alongside model size and data. The fitted laws predict …
A new arXiv paper (arXiv:2610.00705v1) develops a meta-multi-agent reinforcement learning (meta-MARL) framework that models multi-agent reinforcement learning problems as Markov games and enables rapid interactive policy…
SaharaLabsAI researchers introduced ReLiveGym, a diagnostic evaluation environment that tests long-lived LLM agents over simulated weeks of chronologically replayed real-world news, market, and social-media streams, per …
A new arXiv paper (2609.38360v1) proposes Student-COnditioned Updates of the Teacher (SCOUT), a co-training framework that adapts the teacher model to student-generated prefixes in on-policy distillation (OPD). The autho…
A new arXiv paper (2609.38383v1) presents a conditional energy-based model that learns temporal log-density ratios from random exploration, without action or reward labels, to enable long-range planning. Trained via nois…
A symbolic Lean 4 pipeline called LeanPolish released 33,402 accepted local edits and 65,596 same-state failed attempts to study what language models learn from verified proof-edit supervision, according to the arXiv pap…
A controlled CarRacing study found that Disagreement-Regularized Imitation Learning (DRIL), which converts disagreement among cloned policies into a reinforcement-learning reward, improved over the strongest behavior-clo…
Researchers submitted FlashDiffusion, a matrix-free method for diffusion maps and kernel spectral decomposition, to arXiv on 18 Sep 2026. FlashDiffusion evaluates dense Gaussian kernel blocks in fused GPU tiles and coupl…
A new arXiv paper (2610.00012v1) introduces FedCausalCompose, a causal world-model framework that shows causal structure helps modular LLM agents only when cross-module interfaces are both statistically identifiable and …
Researchers posted arXiv:2610.00061v1, introducing GAP-DPO (Geometry-Aligned Preference DPO), an iterative algorithm that selects preference pairs by aligning their gradient direction with expected user utility gradients…
Researchers introduced the Conformal Adversarial Generative Ensemble (CAGE), a forecasting framework that combines generative modeling, adversarial discrimination, and conformal prediction to reduce the influence of extr…