Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
MIT researchers have developed Tilted REWeighting (TREW), a method that fixes privacy leaks in AI class unlearning by adjusting the model's distribution to mimic a scratch-trained model. On CIFAR-10, TREW reduces the per…
A new AI-powered framework called InqEduAgent uses a Gaussian process to analyze students' cognitive and evaluative traits, then pairs them with compatible study partners to improve collaborative learning outcomes. The s…
A new AI-driven dual-attention deep reinforcement learning system reduces scheduling delays in prefabricated construction by reaching within a 4% margin of constraint-programming references, outperforming traditional dis…
A new study proposes an active policy selection method for offline-to-online reinforcement learning that uses upper-confidence bounds to allocate limited online interactions between evaluation and fine-tuning, consistent…
A new AI model called GenGNN challenges the need for complex Graph Transformers in discrete graph generation, achieving up to 5x faster inference times with a simpler message-passing approach. Developed by researchers, G…
A new framework automates quantum circuit design using graph-based Bayesian optimization and a graph neural network surrogate, achieving competitive classification accuracy on cybersecurity data with reduced circuit comp…
Researchers have developed a transformer-based imitation learning framework for soft robotic guidewire navigation, achieving an 83% success rate on unseen vascular geometries and 75% on a patient-derived geometry. The sy…
Researchers have introduced identity bridge supervision to overcome the curse of two-hop reasoning in large language models, enabling out-of-distribution generalization even in simple one-layer transformers like Emb-MLP.…
A developer has built a searchable chat interface for Stanford's CS229 machine learning course notes, allowing users to explore the material with citations back to the original source. The tool, hosted at chat.pageindex.…
Parse for Artisans, a hosted parsing API with a Laravel-native SDK, enables Laravel 13 apps to convert uploaded PDFs, scans, .doc files, and emails into Markdown for AI-powered document Q&A. The pipeline uses Postgres wi…
Adversarial World Modeling (AWM), a new framework for training autonomous vehicle planners, uses a min-max game and multi-agent self-play to enhance performance in rare, safety-critical traffic scenarios. Tested on nuPla…
Diachronic Sample Integration (DSI), a new approach for deep generative models, reduces tail-estimation errors in rare event prediction by aggregating samples across training checkpoints, outperforming standard methods i…
Researchers have introduced the Causal Discovery Foundation Model (CDFM), a unified framework for zero-shot structural causal inference that consistently outperforms traditional algorithms across diverse domains. By trea…
RepTran, a new AI repair method targeting Transformer feed-forward networks, achieved a 74.7% average repair rate across 18 fault benchmarks from CIFAR-100 and Tiny-ImageNet, outperforming random selection and the Arachn…
A University of Lausanne study found that an explainable machine learning pipeline assessing mental health risks in students produced stable results that reflected data construction rather than actual outcomes. When trai…
A developer provides a technical guide for compiling llama.cpp on Debian 12/13, covering native builds and cross-compilation from x86_64 to ARM64. The guide includes steps for enabling OpenBLAS and verifying binary archi…
AMID, a new multi-agent framework for automating medical imaging model development, outperformed existing general-purpose machine learning engineering systems across 20 medical imaging challenge tasks, approaching or mat…
Researchers have identified that the real cause of training instability in reinforcement learning with flow-matching policies is not iterative action generation but the vanilla sampling strategy, and they have introduced…
A study testing 15 foundation models for mammography across 12 out-of-distribution datasets found that models like Mammo-FM and MaMA showed promise but struggled with generalization, while DINOv3, a vision-only model, pe…
Researchers have developed HUMA (Hybrid Understanding for Multi-modal social Navigation), a new architecture that combines reinforcement learning with vision-language models to improve robot navigation in human spaces. E…