Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
MIT researchers have developed a new auditing method called Gaussian probing that can detect whether an AI model has been fine-tuned to generate child sexual abuse material (CSAM) with 100% accuracy without ever producin…
USC assistant professor Ibrahim Sabek has received a $627,250 NSF CAREER Award for a five-year project titled "Toward Quantum-Augmented Database Systems" that aims to integrate quantum computing into database engines to …
A developer has built a Fashion-MNIST neural network trainer in pure SQL using xarray-sql, demonstrating that matrix multiplication and backpropagation can be expressed as joins and aggregations. The three-layer MLP with…
A developer compiled 14 articles on browser testing that address modern challenges like layout shifts, CSS view transitions, AI-generated interfaces, and the limitations of green CI signals. The articles emphasize that w…
Tessen.ai, a strategy-grading engine, tested 13 widely taught trading strategies on six years of crypto data across 10+ major pairs using strict out-of-sample validation and real costs. All 13 failed to pass all five pas…
A developer implemented a neural network in SQL using the xarray_sql library, achieving a speedup of approximately 1.8x on Fashion-MNIST by skipping zero-valued pixels during layer-0 contraction. The project, shared on H…
An Android developer explains how to implement always-on audio AI using a Digital Signal Processor (DSP) to avoid battery drain and overheating. The post details the hardware paradox where CPUs are too power-hungry, GPUs…
A new Python interview guide, 'Python Under Pressure', written by a software engineer with 15+ years of experience building ML/AI systems, aims to close gaps in candidates' knowledge by covering CPython internals, perfor…
A developer has created PVM (Persistent Vector Memory), an open-source Python library (~800 lines) that gives LLMs long-term memory without API keys or GPUs. It uses temporal weight decay, delta rule calibration, and fuz…
A developer built MargIQ, a tool that analyzes recurring AI workflows to determine which tasks can use cheaper models without sacrificing reliability. MargIQ observes production traffic, identifies workflows, and recomme…
Distillation has become a hot topic in AI post-training, as highlighted in a recent article by Sergio Paniego that traces the technique's history ahead of a class on distilling open models with TRL. The latest GLM-5.2 re…
A developer built StudyMate AI, a voice-first tutoring system for ICSE/CBSE students that tracks cognitive state over time using a behavioral graph in Neo4j. Unlike typical AI tutors, it classifies students into four sta…
A USC-led study used AI to analyze biomarkers including sweat, brain signals and eye movements to classify depression and suicidal ideation, offering a potential objective testing method for mental health conditions. The…
Apple's new SpeechAnalyzer engine, introduced with iOS 26 and macOS 26, beats OpenAI's Whisper Small on clean English read speech with a 2.12% word error rate on LibriSpeech test-clean versus Whisper Small's 3.74%, accor…
Amazon SageMaker AI launched a UI for generative AI inference recommendations in SageMaker AI Studio, providing a low-code no-code experience that guides users through preset use-case profiles, visual comparisons, and on…
Four months after Google Research and DeepMind published TurboQuant, a training-free algorithm suite claiming 6x memory reduction for KV cache, the promised official code has not shipped. Instead, the community has produ…
A new research paper argues that deep learning algorithms fail to learn from unfiltered experience because they are designed for human-curated datasets. The authors demonstrate that standard optimizers like SGD and Adam …
TormentNexus' AI skill registry has surpassed 5,776 reusable modules, with top categories including code review, Terraform generation, and database migrations. The code review skill family alone comprises 1,072 modules a…
A new unified scaling law for quantization-aware training (QAT) models quantization error as a function of model size, training data volume, and quantization group size, based on 268 experiments. The study finds that qua…
Roboflow published a tutorial on automating surface defect detection using a custom RF-DETR object detection model and Roboflow Workflows, enabling quality teams to instantly classify product health and route defective i…