Neural Networks

A developer has released NeuroAnswer v2, an MCP server that allows Claude to navigate and query the MICrONs Minnie65 connectome dataset containing 200,000 neurons and 500 million synapses using plain English. The system …

Facebook AI's FAISS library enables billion-scale similarity search by converting data into vector embeddings and using approximate nearest neighbor techniques. The system combines Inverted File (IVF) partitioning to ski…

Canonical engineer Asa Mirzaieva demonstrated how to deploy optimized AI models on Renesas RZ/V series hardware using Ubuntu Core 26 and the Dynamically Reconfigurable Processor for AI (DRP-AI). The workflow enables deve…

Jeff Bezos, Lux Capital, and Google Ventures have invested $500 million into Flourish, a New York neuro AI startup founded by Thomas Reardon and Rob Williams, betting that the next AI breakthrough will come from studying…

Jeff Bezos is funding Flourish, a neuro AI company co-founded by former Amazon executive Rob Williams and neuroscientist Thomas Reardon, to develop a synthetic intelligence system that mimics the human brain's power effi…

Researchers have developed DiffusionBlocks, a method that trains neural networks by dividing them into blocks and training only one block at a time, reducing memory requirements to roughly 1/B of standard end-to-end trai…

The California two-spot octopus genome reveals no evidence of whole-genome duplications, but shows massive expansions in protocadherins and zinc-finger transcription factor gene families previously thought unique to vert…

Nvidia has acquired Kumo AI, a four-year-old startup that builds predictive foundation models for business data, for at least $400 million according to The Information. The three Kumo co-founders moved to Nvidia last mon…

Researchers have extended refusal steering methods to Mixture-of-Experts (MoE) large language models, demonstrating that steering vectors can effectively suppress safety-aligned refusal behavior in these architectures. T…

Researchers have developed a methodology for the inverse design of critical experiments using deep neural network surrogate modeling and gradient optimization to maximize neutronic similarity to target nuclear technologi…

Researchers have introduced DLLG (Dynamic Logit-Level Gating), a framework that learns token-level fusion weights for combining multiple specialized large language models without requiring token-level labels or expert re…

Researchers introduced Stein kernelized molecular dynamics (SKMD), an enhanced sampling method that uses interacting particle dynamics to acquire informative training configurations for active learning of machine learnin…

Researchers have developed LiftQuant, a framework enabling continuous bit-width control for large language models by using a "lift-then-project" mechanism that approximates low-dimensional weight vectors from a higher-di…

Researchers introduced SDPG, a self-distilled policy-gradient framework that combines group-relative verifier advantages with normalized standard deviation and full-vocabulary on-policy self-distillation to improve reinf…

A systematic study of query, key, and value (QKV) projection variants in Transformers found that sharing the key and value projections (Q-K=V) performs on par with or better than the standard three-projection approach ac…

A new pseudospectral theory for block-triangular Jacobians in coupled gradient descent proves that transient amplification before convergence can be arbitrarily large even when asymptotic stability is guaranteed, with th…

The IEEE P3109 draft standard defines a new family of parameterized binary floating-point formats designed for machine learning, allowing efficient value representation in few bits with adjustable width, precision, signe…

Researchers introduced Learnable Rank LoRA (LR-LoRA), a parameter-efficient fine-tuning method that allows the adapter rank to be learned during training rather than fixed, enabling the optimizer to determine the appropr…

Researchers introduced XSSR, a cross-domain self-supervised framework that selects representative medical images for annotation, reducing labeling effort while maintaining segmentation accuracy. The method achieved 99.3%…

Researchers have developed a training-free diffusion model that generates new images matching the internal patch structure of a single reference image, eliminating the need for computationally expensive neural network tr…