Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
NexusMem, a local-first persistent memory engine for AI coding agents such as Claude Code, Cursor, and MCP-based agents, records git history, shell commands, docs, and conversation transcripts into an on-disk SQLite data…
Startups are shifting from precision-based robotics to AI-driven perception to teach robots to fold laundry, using neural networks trained on human movement to handle unstructured objects. Mastering laundry is seen as a …
TileRT's persistent engine on NVIDIA GPUs achieves up to 500 tokens/s/user on the InferenceX GLM5 FP8 744B benchmark on a single B200 decode server, approximately 3× faster than GB300 NVL72 running traditional inference …
A developer has reverse-engineered Apple's private APIs to train neural networks directly on the Apple Neural Engine (ANE), bypassing CoreML's inference-only restriction. The project, which uses the _ANEClient and _ANECo…
A developer used MCP's Resource primitive to fix a LangGraph-based support bot that lost conversation context, enabling efficient knowledge graph updates and retrieval across tool calls. The approach centralizes device i…
AI pair programming assistants like Cursor and Claude Code are shifting from simple autocomplete to full task delegation, enabling developers to automate repetitive coding tasks and focus on complex problems. These assis…
TensorFlow's XLA compilation path silently converts tf.debugging.Assert and tf.Assert operations into no-ops, disabling runtime checks in jit_compile=True functions. This deliberate behavior, implemented in tensorflow/co…
Researchers found that safety alignment in Diffusion Large Language Models (DLLMs) is sparse and transferable, enabling attacks that increase attack success rates from 2.6% to 73.8% on LLaDA and from 1.9% to 86.6% on Dre…
Researchers introduced Graph Machine, an architecture with two explicit edge-based mechanisms—edge-augmented attention and edge-centric referral—that outperforms Transformer baselines on Sudoku reasoning tasks, with abla…
Researchers from an undisclosed institution present an unsupervised method that extracts linguistic metaphors from a corpus and uses structured clustering to identify conceptual metaphors, revealing framing differences a…
Researchers propose Stage-Guided Per-Step Optimization (SGPO) for diffusion models, which assigns stage-specific objectives during denoising to address reward sparsity and temporal objective mismatch in reinforcement lea…
Researchers propose TransSLR, a lightweight Temporal Transformer Encoder trained from scratch on 64-frame normalized pose sequences, achieving 80.39% accuracy on the CASL-W60 benchmark for Central African Sign Language r…
Researchers introduced EB-CaP, a test-time adaptation method for video facial expression recognition that personalizes class prototypes per target video using an energy-based model and CLIP embeddings, outperforming stat…
Researchers at arXiv introduced new minimax-optimal sample complexity bounds for learning epsilon-optimal robust policies in average-reward Markov decision processes under total-variation uncertainty. The study, publishe…
Researchers from an unnamed institution propose AgentPatch, a training-free coarse-to-fine repair framework that merges agentic multimodal large language models (MLLMs) into a single generalist checkpoint, addressing asy…
Researchers propose NTDH, a complex-reasoning framework for comprehensive affective analysis that unifies sentiment and emotion tasks under one output interface and a verifiable reward trajectory. Trained on 16,302 recor…
Researchers at an undisclosed institution trained a multi-task neural network to recover lesion parameters from aphasic picture-naming error profiles in LLaVA-Vicuna 13B, finding that modification percentage and noise si…
Researchers at arXiv introduced Spherical Soft-Masking (S-SM), a drop-in replacement for linear interpolation in Masked Diffusion Language Models (MDLMs), after finding that the embedding space is hyperspherical with a n…
Researchers introduced Task-Expert-Aware Supervision (TEXAS), a method that identifies task-relevant experts in Mixture-of-Experts (MoE) language models by comparing activations on successful versus failed instances and …
A new arXiv study (2608.06607v1) finds that pre-inference routing can cut document-extraction costs by 31-33% on receipts and 77% on degraded ad-buy forms while keeping F1 within 0.02 of always using the large model, but…