Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Chinese robotics company Astribot released a video demonstrating its new 'SmoothRL' technology, a reinforcement learning method that enables robots to learn from their actions in real-world environments, bypassing simula…
Researchers introduced speculative decoding, an algorithm that accelerates inference from large autoregressive models like Transformers by computing several tokens in parallel without changing outputs. Demonstrated on T5…
Researchers introduced Sarathi-Serve, an LLM inference scheduler that uses chunked-prefills and stall-free scheduling to improve throughput-latency tradeoffs, achieving 2.6x higher serving capacity for Mistral-7B on a si…
An MSc Computer Science student at the University of Nairobi is seeking arXiv endorsement for a paper on a protocol-aware adversarial attack framework for MQTT-based intrusion detection systems. The framework achieved 6.…
Alphabet Inc.'s Google and Cathay Pacific Airways Ltd. reported Monday that a pilot project using Google's AI algorithms on several dozen Cathay routes in late 2025 achieved a 40% decline in the warming impact of contrai…
EXAONE Finance, a financial forecasting model from LG AI Research, ranks first across all three FinVerse evaluation tiers, according to a paper posted on arXiv (2609.04239). The model uses a novel attention-free architec…
A systems-security case study of a two-node split-LLM training system found that the returned gradient leaks private data, nullifying decoy defenses, with the pattern of zero gradients revealing real rows on every frame …
Timothy Gordon released Engrim 1.3.0 on September 7th, an open-source tool that gives Claude Code, Cursor, Windsurf, and Google Antigravity a shared local SQLite memory layer to preserve project decisions across sessions…
A developer has released a large Word2Vec model trained on 13 billion words of Russian fiction, available on Hugging Face under the MIT license. The lemma-based CBOW model with 300-dimensional vectors captures literary s…
Kubernetes 1.37 “Garhwal,” released August 26 with 67 enhancements, now natively supports scaling workloads to zero replicas via the Horizontal Pod Autoscaler (HPA) in beta, eliminating the need for KEDA in basic queue-c…
CauterRule, an open-source tool that converts repeated agent failures into standing rules, released v0.1.0 after a field test comparing four models across 394 trajectories. The test found that even the strongest model, m…
A developer teardown explains how AI LinkedIn headshot generators work, detailing the training pipeline that fine-tunes a latent diffusion model on 10 to 20 selfies using DreamBooth and LoRA. The process involves persona…
Software engineer Shane Hobson published a guide explaining how large language models (LLMs) work under the hood, aimed at fellow software engineers. The guide walks through the LLM architecture step by step, from prompt…
Researchers introduced LETHE (Latent-parameter Evolution with Temporal Hierarchical quasi-Equilibrium), a self-referential sonic-oblivion system implemented in SuperCollider that adapts GAN-inspired architecture without …
A new study from arXiv (2609.04336v1) introduces MedProb, a lightweight probing framework that predicts multiple-choice medical visual question answering (Med-VQA) answers from frozen vision-language model (VLM) represen…
Researchers introduced SVDtrunc, a two-step block-level compression scheme for Diffusion Transformers (DiTs), which outperforms competing approaches across GenEval, HPSv2, and DPG benchmarks while retaining near-full per…
Researchers propose a Fourier auxiliary head to address physical representation laziness in JEPA-style latent world models, improving planning success rates in dynamic environments. The method, tested on the LeWorldModel…
Researchers introduced AdaptVPR, a route-aware generative augmentation framework that constructs same-place hard positives to improve Visual Place Recognition (VPR) robustness against domain shifts. The framework, which …
Researchers introduced FailSAE, a framework using Sparse Autoencoders (SAEs) for interpretable failure prediction in vision-language models like CLIP, outperforming baselines in experiments. The method employs a three-st…
Researchers introduced Reflection-Aware GRPO (RA-GRPO), a reinforcement learning framework for diffusion-based visual generation that improves preference alignment by incorporating backward reflection into forward genera…