Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A developer detailed the architecture for building a real-time object detection and tracking app on Android using Kotlin and YOLO-style models. The approach separates camera capture, inference, tracking, and rendering, a…
OpenAI's GPT-2 medium weights achieved the highest IFT score of 42.43 with only 2 IFT epochs, while a JAX model with no dropout and no MHA bias scored 21.45 with 5 epochs, and a JAX model with dropout scored 17.17 with 7…
A developer built a hybrid retrieval-augmented generation (RAG) support assistant for a fictional B2B SaaS platform, 'Helix,' and found that hybrid search underperformed vector-only retrieval. Controlled before/after tes…
AdBuster Team, based in Dublin, released AdBuster 2.0 PRO, a Windows application that uses ML/AI and its proprietary CEPA logic to stabilize TV volume spikes from commercials in real time. The free trial is available for…
A solution writeup for the ECML 2026 Flatland Real-World Baselines Challenge achieved rank 1 on the leaderboard, using a SIPP train dispatcher to route trains on a 120x150 grid with 28 stations across 35 scored episodes.…
Silicon Valley has adopted 'hill-climbing' as its favorite new jargon, according to a Wall Street Journal article. The term, borrowed from optimization algorithms, is now used by tech executives and engineers to describe…
Castform, the AI training platform from Benchmax, has added supervised finetuning (SFT) support, enabling developers to post-train models via next-token prediction on demonstration data. In a demonstration, a Qwen3.5-4B …
LangChain, the AI startup with $125M in Series B funding from IVP, Sequoia, Benchmark, CapitalG, and Sapphire Ventures, is hiring a Research Engineer for its LangSmith Engine team in New York, NY or San Francisco, CA. Th…
A developer has released a clone-and-run launcher that runs LTX-2.5 text-to-video generation locally on Apple Silicon in a single command, with audio support and no need for ComfyUI or CUDA. The launcher achieves a 16x s…
Researchers introduced Faraday, a 27B-parameter AI Scientist agent that outperforms Claude Opus 4.8 and GPT-5.5 on replicating research figures, trained via long-horizon RL on the Replica task suite of 310 tasks from 100…
Hugging Face hosts over 100,000 models and thousands of datasets as of 2024, making it the central hub for open-source machine learning, according to a guide on where AI developers should seek help. The guide recommends …
A new open-world perception system for driving scenes safely handles unknown road objects by using a hierarchical taxonomy, abstraction fallback, and segmentation cross-validation, as described in a reference implementat…
A new contract-grade verifier with twelve adversarial gates found that 39.5% of 2,638 machine-generated GPU kernels accepted as correct by a public system's harness are broken beyond any tolerance argument, and 62.1% car…
A new educational exercise demonstrates that the Discrete Fourier Transform (DFT) can be calculated by hand as a series of matrix multiplications, revealing the mathematical foundation behind signal processing and deep n…
Developer egnaro9 released evalmut, an open-source mutation testing tool for LLM evaluation suites, available via pip install evalmut and on GitHub under an MIT license. The tool injects 18 provenance-gated defects into …
VLLM outperforms Ollama by nearly 20x in throughput at high concurrency, according to benchmark tests running Llama 3.1 8B on an NVIDIA A100 40GB, with vLLM peaking at 793 tokens per second versus Ollama's 41 tokens per …
Researchers from an undisclosed institution introduced Hybrid Domain Knowledge Fusion (HDKF), a relighting-specific training framework that fuses physics, reflectance, and realism priors from synthetic, One-Light-at-a-Ti…
Researchers propose HorizonRelight, a diffusion transformer-based framework that reframes long-horizon video relighting as temporally conditioned latent domain translation, using masked target-domain self-conditioning an…
A developer has built NexusMem, a local-first memory engine for AI coding agents that indexes git history, shell commands, project docs, and assistant transcripts into a SQLite database, serving ranked context over MCP o…
Amazon Web Services (AWS) published a technical guide on designing custom reward functions for multi-turn reinforcement learning (RL) with Amazon Nova Forge, emphasizing that the reward function determines what the model…