Neural Networks

A neural network design called forking-sequences reduces forecast volatility in time series forecasting by encoding and decoding the entire time series across all forecast creation dates in a single forward pass. Studies…

CoCoT, a contrastive-pretrained EEG model with multiscale temporal convolution layers and Transformer encoders, outperforms state-of-the-art reconstruction-pretrained models on benchmark tasks, challenging the dominant a…

Pattern-aware graph neural networks that encode missingness patterns improved balanced accuracy by 17% and F1-macro scores by 22% across seven UCI datasets, according to a study by researchers exploring novel data imputa…

Researchers propose LieBN, a framework for Riemannian Batch Normalization over Lie groups that leverages left- and right-invariant metrics to normalize manifold-valued sample distributions. The framework is instantiated …

A new study from researchers at Universitat Politècnica de Catalunya presents a Variational Graph Autoencoder (VGAE) that detects power flow solution feasibility, using the IEEE 118-bus case to assess the validity of sol…

Researchers propose PRecG, a pipeline that uses graph neural networks and rhetorical role segmentation to improve legal precedent retrieval by capturing nuanced legal meanings within documents. The approach decomposes ju…

Researchers at arXiv analyzed eleven pretrained GPT-2-style checkpoints and found shared depth trends in weight spectra, but their initialization schemes imitating component-wise magnitudes and spectral profiles did not …

A new theoretical framework from researchers reduces the adversarial robustness problem for multilayered perceptrons (MLPs) to a lattice traversal problem, introducing both sound and complete interval certifications. The…

A new study from arXiv provides exact solutions for how linear representations emerge in neural networks during training, a process called abstraction. The researchers found that data and target geometry jointly determin…

A new transformer architecture achieves 78% legibility in feed-forward operands and 50% in attention value channels by training operators to be crisp, bounded, and named as fuzzy set operations, researchers report in arX…

Vision Transformers (ViTs) robustly encode Gestalt-like figure-ground cues such as surroundedness and convexity from natural images, according to a study evaluating 25 ViTs with supervised and self-supervised training. R…

Cognitive neuroscientists at MIT's McGovern Institute for Brain Research have shown that people can perform logical reasoning tasks even with severely impaired language abilities, challenging the long-held theory that la…

Researchers from IT University of Copenhagen, Sakana AI, and Autodesk have developed a system of hundreds of simple cubic bricks that each run identical local neural networks and communicate only with physically connecte…

Researchers studying neural networks discovered that individual neurons do not represent single concepts but instead respond to multiple unrelated features, a phenomenon called superposition. This finding challenges the …

Microsoft researchers introduced ResNet in 2015, a neural network architecture that uses skip connections to solve the degradation problem in deep networks. By allowing layers to learn residual mappings instead of full t…

In 2012, Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton published the AlexNet paper at NIPS, introducing a deep convolutional neural network architecture that revolutionized computer vision by winning the ImageNet …

A developer published a Keras-based implementation of the LeNet-5 convolutional neural network architecture, originally proposed by Yann LeCun in 1989 for handwritten digit recognition. The implementation uses functional…

Kexin Wang's Funguy project uses a laser diode to trace outlines on agar gel, controlling the growth of photophobic fungi to create art. The system relies on a computer model combining a temporal convolutional neural net…

A developer's MS thesis appendix explores Bayesian neural networks as a solution to underspecification in deep learning. The post covers methods like Monte Carlo dropout, Bayesian last layer approximation, and stochastic…

AI systems are generating a dynamic, reactive open world for GTA 6, where weather, NPC interactions, and environmental damage persist and evolve without manual scripting. The technology, which already exists, could make …