Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A tutorial demonstrates how to build, track, compare, and register scikit-learn pipelines that integrate large language models using Scikit-LLM and MLflow, with code for logging multiple pipeline versions across differen…
Desert Ant Labs, a European AI lab founded by the team behind the video app Detail, launched 18 on-device models for audio, vision, and text, including Voz, a transcription model 4.7x faster than Whisper, and Clear, a 9M…
Radical Numerics has post-trained its Omnii genome model to design personalized mRNA cancer vaccines, ranking neoantigens and writing mRNA cassettes from tumor mutations and MHC alleles, but has not yet presented human, …
Databricks has developed KARL, a retrieval-augmented generation agent that uses reinforcement learning to stop searching when it has enough information, matching the performance of Claude Opus 4.6 while costing 33% less …
Ludwig, a declarative deep learning framework hosted by the Linux Foundation AI & Data, announced support for Python 3.12, PyTorch 2.7+, Pydantic 2, Transformers 5, and Ray 2.54, enabling users to train, fine-tune, and d…
Hugging Face launched 'ML Intern,' an AI assistant built into its chatbot that lets users run machine learning experiments without any ML expertise.
A developer detailed the construction of a vector search engine from scratch using the HNSW algorithm in Python, explaining the layered graph structure and query-time complexity reduction from O(n) to O(log n). The imple…
An educational article explains unsupervised machine learning, a technique that finds patterns in unlabeled data without human-provided answers. It details clustering algorithms such as K-Means and hierarchical clusterin…
OpenWAM, an open and modular framework for systematic world-action model pretraining, was introduced to address the monolithic nature of existing systems by separating the generative backbone, visual representation, arch…
Fast Company's eighth annual Best Workplaces for Innovators list, announced today, recognizes Lenovo as a finalist in both the AI, Automation, and Machine Learning Excellence and North America categories for the third co…
Stanford University professor Chris Piech announced a new online course, Probability for AI, starting October 9th with applications due at the end of September, as part of a community service project to teach machine lea…
Marigold V2, a diffusion transformer-based model for monocular depth estimation, revisits the approach to improve generalization to out-of-distribution inputs, addressing a persistent challenge in the field with applicat…
Researchers proposed cross-model KV cache transfer, a method that reuses a source model's key-value cache to skip prefill when swapping between different-sized models in a family, and found a closed-form linear mapping r…
FuturePresentLabs released tts-jarvis, an Apache 2.0-licensed LoRA adapter for VoxCPM2 text-to-speech, on Hugging Face, along with a Rust-backed dataset studio called Tape Sampler that uses an edit decision list to manag…
Researchers introduced RelightFormer, a feed-forward generative transformer that performs multiview object relighting by integrating multi-view cues to overcome the limitations of traditional inverse rendering and single…
Timur, a researcher, is requesting an arXiv endorsement to submit a paper on S2S, a physics certification method that applies 16 biomechanical laws to validate IMU/EMG sensor data before robot training. The method report…
Bitmovin's 10th Annual Video Developer Report, based on responses from 486 video professionals in 94 countries, finds that only 16% now consider sub-five-second delivery as low latency, down from 60% five years ago. The …
Goodfire used Ai2's fully open post-training stack — the Dolci preference dataset, Olmo 3 intermediate checkpoints and training recipes, and the OLMES evaluation suite — to predict behavioral changes before a full traini…
A newly disclosed discriminative model that combines time-domain and frequency-domain features with a cosine similarity loss function improves detection of unknown electromagnetic waveforms by 10% over models lacking the…
Researchers propose SQS, a Bayesian method that unifies pruning and quantization for deep neural network compression by modeling weights with sparse quantized sub-distributions, achieving higher compression rates while p…