Neural Networks news and analysis on Web Pulse: 740 curated articles tracking the latest Neural Networks developments, tools, and research, updated continuously from vetted sources.
A developer explores the concept of an internal 'workspace' in large language models, analogous to a reasoning blackboard, which emerges from training rather than being explicitly designed. Experiments show that manipula…
Researchers propose that group theory, particularly concepts of symmetry, representation spaces, and invariance, provides a useful framework for understanding how Transformer architectures learn and organize information.…
Researchers found that formal robustness certificates for embedded neural-interface models can pass even when task accuracy collapses, with EEGNet classification accuracy dropping 25.7% under attack while the certificate…
A new audit of grokking representation metrics reveals that effective rank measurements at the grokking transition overstate converged values by 3-5x on MLPs and 1.3-1.5x on transformers, with compression lagging accurac…
Researchers introduced a method for extracting rail tracks from classified 3D point clouds using a fully convolutional recurrent neural network trained on synthetic data. The approach rasterizes points, applies the netwo…
Researchers at the GRAPES-3 experiment developed a graph neural network (GNN)-based pipeline for cosmic ray energy reconstruction, achieving improved accuracy and automated feature extraction on sparse detector arrays. T…
Researchers introduced STST-JEPA, a self-supervised transformer for EEG analysis, pretrained on 47,703 sessions from ages 5-81. The model achieved a mean absolute error of 3.06 years in age regression and ranked first on…
Researchers propose STAGformer, a Spatio-Temporal Agent Graph Transformer for micro mobility demand forecasting in bike-sharing systems. The model achieves efficient global modeling with linear computational complexity b…
Researchers propose MiLSD, a micro line-segment detector designed for resource-constrained devices such as microcontrollers, achieving competitive accuracy within a one-megabyte memory budget. The detector uses a compact…
A new study analyzing seven pretrained models reveals that the positional encoding scheme in transformers determines the spectral properties of attention heads, with RoPE inducing rotational spectra and absolute/ALiBi sc…
Researchers introduced CoFINN (Conservation Flux Informed Neural Networks), a physics-informed deep learning framework that embeds finite-volume conservation physics into CNN training for compressible flow prediction. Te…
Researchers developed a physics-guided machine learning framework that integrates domain knowledge into deep learning models to predict fuel density for wildfire management, outperforming purely data-driven approaches in…
Researchers introduced a Siamese Neural Network framework to identify phase transitions in 3D percolation models using only 22 labeled probability points from non-critical regions. The method located percolation threshol…
Experiments building on a mathematical theory of transformers reveal that the architecture drives tokens to cluster and collapse through layers, but trained weights learn to resist this clustering, enabling the model to …
Researchers solved BlueDot's TAIS Puzzle #1 by identifying a nonlinear representation of the 'country' feature in a five-layer MLP, hidden as an XOR with 'food' at layer h2. They used Distributed Alignment Search to prov…
A writer argues that free will can be modeled as a learned, context-dependent parameter in machine learning, using the VAE's mean and standard deviation as a better representation than binary or global parameters like te…
A developer explains how Quantization-Aware Training (QAT) enables deploying deep learning models on Android without sacrificing accuracy, addressing the 'Quantization Cliff' problem of post-training quantization. The gu…
The Premium Machine Learning Artificial Intelligence Super Bundle, a twelve-course digital toolkit for learning AI and data science, is on sale for $37 via the Techdirt Deals Store, which is powered by StackSocial.
Anthropic researchers discovered a 'global workspace' inside language models like Claude, a small region of internal representations that behaves like the brain's conscious access hub. Using a new interpretability tool c…
A study in Nature reveals that individual brain cells in the human cortex act as specialized building blocks for spoken language. Researchers recorded electrical activity from 579 neurons in eight participants during nat…