Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Google released EmbeddingGemma 2 on Tuesday, an open-source on-device model with 740 million parameters (270 million for text-only workloads) that adds code, image, video and audio support beyond its text-only predecesso…
A developer built Podium, a speech-delivery analysis tool that force-aligns a speaker's reading of a public-domain speech (JFK, Reagan) word-by-word against a reference delivery using torchaudio's MMS_FA wav2vec2 CTC ali…
Google released EmbeddingGemma 2, an open embedding model with 740 million parameters that converts text, images, video, audio, and code into vectors and runs on-device using about 191 MB of RAM. According to Google, Emb…
An AI agent named Alfred, working for developer Brooks Moore, built Lakefront Window, an open-source tool that uses a PriorLabs-TabPFN model trained on 2015–2023 sensor data to predict good outdoor hours on Chicago's lak…
A developer demonstrated that 4-bit GGUF quantization of Mixture-of-Experts models can match full-precision quality when sensitive tensors are kept at higher precision, reporting MMLU-Pro 87.65 for a mixed-precision UD-Q…
Google launched EmbeddingGemma 2, a 740-million-parameter open embedding model under the Apache 2.0 license that maps text, code, images, video and audio into a single shared space and runs on-device. The model, built on…
NVIDIA's DOCA GPUNetIO has evolved from a standalone GPU-centric packet-processing framework into a unified GDA-KI foundation now integrated across NVSHMEM, NCCL, Aerial 5G SDK, UCX/NIXL, NVQLink with the Holoscan Sensor…
A developer built Eyes Up, an offline, privacy-first bird-call companion that records 30 seconds of ambient audio, classifies it on-device with Cornell's BirdNET V2.4 TFLite model, and generates a grounded two-sentence f…
A developer built a small machine-learning classifier to filter political fundraising and advocacy email from a personal inbox, using a 100-email evaluation set split evenly between 50 political and 50 nonpolitical messa…
A research paper finds that general-purpose large language models already function as Jev-style decision models, returning categorical probability distributions over predefined options without generating free-form text, …
A developer built Trail Notebook, an offline field journal that runs BirdNET bird-call detection and a local Gemma 3 4B model to turn trail recordings into Markdown journal entries with species tables, confidence scores,…
A study posted as arXiv:2608.24058 used a Wizard of Oz setup — a human simulating a personalized ML model — to test whether users can define their own sensing phenomena instead of choosing from preset categories, with pa…
Google released EmbeddingGemma 2, a 740-million-parameter open-weight multimodal embedding model licensed under Apache 2.0, on October 6th, mapping text, code, images, video and audio into a shared 768-dimensional space …
Google DeepMind released EmbeddingGemma 2 on October 6, 2026, a 740 million parameter open-weight multimodal embedding model under an Apache 2.0 license that permits commercial use. The model places text, code, images, v…
Google announced EmbeddingGemma 2, a 740 million parameter multimodal embedding model built on the Gemma 4 architecture and released under an Apache 2.0 license, alongside a new Mac notetaking app called Google AI Edge F…
Parsec AI released TOD, a 12B Gemma-based model and API for decision-making in LLM pipelines, open-sourced on Hugging Face. TOD scores options for tasks such as ticket routing, intent classification, urgency and severity…
Hugging Face forum users recommended WhisperX, which combines Whisper large-v3 or large-v3-turbo transcription with Pyannote 3.1 diarization, for transcribing a one-hour two-person English podcast with casual overlapping…
Researchers propose iADD, a method for reinforcement-learning post-training of diffusion models that aims to improve alignment with reward functions without sacrificing output diversity, addressing a limitation of Denois…
Google released EmbeddingGemma 2, an Apache 2.0-licensed open multimodal embedding model built on Gemma 4 that maps text, code, images, video, and audio into a unified 768-dimensional space, scaling from 270M parameters …
Google DeepMind launched EmbeddingGemma 2, an open-weight multimodal embedding model with 740M parameters that maps text, images, video frames, and audio into a single unified vector space for local, privacy-first applic…