Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A developer built PlantGuard AI, an open-source plant disease screening tool that fine-tunes a MobileNetV2 model on the PlantVillage dataset to classify 38 leaf conditions and pairs each prediction with a human-reviewed …
A developer measured how fixed-threshold binarization silently destroys OCR accuracy, finding that a 128 cutoff erased an 11 px #999-on-#f5f5f5 footer entirely (0 characters from ImgIng's Professional, Fast and Ultimate …
A developer built Pocket Field Trip, a local-first app that uses the quantized ONNX conversion of the paraphrase-MiniLM-L3-v2 sentence embedding model to match a user's own wording to one of eight short outdoor activitie…
A developer benchmarked 14 OCR preprocessing variants on Chinese text crops using ImgIng's Professional, Fast and Ultimate tiers on an Apple M4 with WebGPU, plus tesseract.js 5 as a control, and found that character erro…
Microsoft released Microsoft-Decision-1, a decision model built on Alibaba Cloud's Qwen3.5-9B and offered through Microsoft Foundry, claiming 83.5 percent accuracy across 36 benchmarks, a 92.2 percent confidence score, a…
A developer published a tutorial demonstrating how to build a binary customer churn prediction model in Python using Pandas and Scikit-Learn, trained on customer demographics, account information, and service usage data.…
Nace.AI has open-sourced Drex 1.5, a 9B decision model that returns a probability for every option in a single forward pass and reads up to 128K tokens. Drex 1.5 scores 58.08 on Decision Index 0.3.1.
A developer released laya-mlx, a native Apple MLX runtime for Laya typed decision models that performs structured routing decisions in 7–14 ms on an M3 Max. The project strips out autoregressive text generation and PyTor…
Strata ran a 125-billion-parameter Qwen 3.8 Flash-Next model locally on an RTX 5090 with 32 GB VRAM, completing a long-answer prompt in 6.6 seconds versus 33 seconds via OpenRouter's cloud service, a roughly 5x speedup t…
A developer published an Apache 2.0 Colab notebook that serves Gemma 4 E2B on a single TPU v5e chip using a pure-JAX engine and compares Google's 4-bit QAT export against an exact repack of the trained quantization grid.…
A developer writing as Mark Vange of Autom8ly re-ran the open-source decision model Kev-4B (4-bit quantized) over 866 questions and found that answers are bit-identical when requests run alone or in batches of two to thr…
GPGPU-Sim 4.0, a cycle-level simulator for Nvidia GPUs running CUDA or OpenCL workloads, is now compatible with the Accel-Sim simulation framework and can run NVIDIA SASS traces generated by NVBit, according to the proje…
Frontier AI models made little progress on InnovationEval, a new benchmark that tests whether AI can independently discover novel machine learning techniques comparable to human researchers, according to early results fr…
An open-source decision model called THX-01 beat TypeSafe Jev on a 2,843-ticket benchmark, posting 98.4% accuracy versus Jev's 97.4%, 98.8 versus 96.2 on messy text, and roughly 10 ms versus 331 ms in latency, according …
GC AI built 38 projects in a 12-hour internal hackathon testing Jev, TypeSafe's new classification model that returns structured evaluations in 0.1 to 0.5 seconds instead of generating text. An engineer's stress tests fo…
Microsoft unveiled Microsoft-Decision-1 on October 9, 2026, a decision-scoring AI model that assigns calibrated probability scores to fixed answer options for routing, classification, prioritization, verification and wor…
Tiny Audio, a speech-to-text system from developer mazesmazes, connects a frozen pretrained speech encoder to a pretrained LLM via a small trainable projector and trains only about 80M parameters for roughly $25, achievi…
Developer osfv released Talus, a 23-million-parameter terrain diffusion model trained from scratch on 45,000 heightmaps generated by the developer's own procedural generator, which runs entirely in a browser and is publi…
HAL-X AI released THX-01, a 322M-parameter open-source non-autoregressive decision model that answers typed questions over documents in a single forward pass, with roughly 10 ms GPU inference and support for 20 languages…
Anthropic handed a result to Columbia's Josh Alman and MIT's Virginia Williams, whose follow-up paper showed 3SUM can be solved in O(n^1.9992) time, breaking the long-held O(n^2) bound, while OpenAI separately released a…