Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers propose MILP-Evo, a closed-loop framework that uses LLM-guided program evolution to automatically design mixed-integer linear programming (MILP) solver logic, such as cut selectors and branching rules, by gen…
A study from arXiv shows that adding lightweight depthwise convolutions to Qwen3 Transformer blocks improves average accuracy on seven downstream benchmarks while adding less than 0.01% parameters. The best results occur…
Researchers propose S2T-RLHF, a sentence-to-token reward decomposition framework that improves training stability in reinforcement learning from human feedback (RLHF) by allocating sequence-level preference rewards acros…
A new study from arXiv models Chain-of-Thought reasoning as a switching dynamical system, finding that reasoning fine-tuning globally reorganizes latent policy states in language models. Across benchmarks and model scale…
A new study on arXiv proposes combining neuron importance with data-aware low-rank approximation for compressing large language models, achieving performance on par with or better than previous state-of-the-art methods, …
A new deep reinforcement learning framework from researchers achieves up to a 20.8% reduction in average end-to-end latency and a 13% improvement in resource utilization in hierarchical edge-cloud computing systems. The …
Researchers have introduced SechKAN, a Kolmogorov-Arnold Network architecture using hyperbolic secant functions, which achieves superior performance over MLPs and other KAN variants on benchmark datasets including MNIST,…
Researchers present SIFT (Self-Improving, Frozen-gate Training), a dynamic classifier service that uses a cheap CPU-bound pipeline with a SPLADE sparse encoder and LightGBM head, escalating low-confidence cases to an LLM…
Researchers propose Dual Attention Residuals (DAR), a method that brings multi-stream interaction into historical retrieval for Transformer models through reciprocal cross-stream addressing. Across dense models from 0.1B…
A new training-time framework called Dependency-Aware Intermediate QA Supervision (DAIS) improves final-answer accuracy on complex reasoning tasks by converting filtered teacher rationales into stage-level QA records tha…
Researchers propose MUX, a method for high-bandwidth reasoning in language models that distills discrete reasoning steps into continuous multiplexed tokens in a latent space, achieving lossless superposition of reasoning…
A hybrid CNN-GLCM-SVM fusion framework achieved 95.0% accuracy, 93.8% sensitivity, and 96.1% specificity in automated four-class cataract severity classification from standard consumer-grade eye photographs, outperformin…
A new training-free framework called Surprise Forcing improves long-video generation by dynamically allocating resources to key visual evidence and skipping unnecessary denoising steps. Developed by researchers and publi…
A new method called Soft-Prompt Privileged Context (SPPC) for on-policy self-distillation in large language models uses a teacher that differs from the student only by a learnable soft prompt, avoiding post-hoc rationali…
BearingNAS, a Hardware-Aware Neural Network Architecture Search framework, enables in-sensor intelligent fault diagnosis for bearings using only a laptop CPU, achieving 99.50% diagnostic accuracy on the LSM6DSO16IS Intel…
A new study from researchers using NVIDIA PhysicsNeMo finds that Monte Carlo Dropout with concrete dropout achieves well-calibrated uncertainty estimates at a fraction of the computational cost of Deep Ensembles for AI-d…
Researchers have introduced Fusion Embedding, a family of models that unify text, image, video, and audio into a single embedding space, enabling cross-modal retrieval without paired audio-visual training data. The first…
A rationale-guided knowledge distillation framework for cross-lingual stance detection outperforms competitive baselines on multilingual benchmarks, according to a new arXiv preprint (2607.18693v1). The method uses Chain…
ExecuTorch's experimental MLX delegate, released in May 2026, enables PyTorch models to run on Apple Silicon GPUs. Testing Qwen3-0.6B on an M1 Max, MLX INT4 achieved 188.9 tokens/s, 4.52x faster than PyTorch MPS BF16 (41…
Researchers propose IGGT4D, a streaming instance-grounded geometry Transformer for online 4D scene understanding that unifies camera motion, scene geometry, and consistent instance features within a single feed-forward f…