Neural Networks

Researchers propose CoANeRV, a coordinate-aware token-space framework for amortized video representation that forms compact video tokens in one feed-forward pass and uses a shared coordinate-conditioned decoder, avoiding…

Researchers propose the kernel risk-sensitive mean p-power based random vector functional link (KRPRVFL) model, which integrates the KRP criterion to replace the standard least-squares objective, improving robustness aga…

Researchers introduced SAGE, an uncertainty-modulated surrogate-gradient mechanism for Transformer-based spiking neural networks (SNNs), which adapts the surrogate-gradient slope during training using normalized self-att…

Researchers propose CutClean, a privacy-aware neural network pruning method that reduces private attribute leakage while increasing sparsity, according to a new arXiv preprint (2608.13773v1). The method uses auxiliary li…

Researchers propose HI-MeshGraphNets (HI-MGN), a hierarchical multi-scale graph neural network that improves mesh-based physics simulation accuracy and efficiency. In benchmarks on three structural and fluid problems, HI…

Researchers propose Batch-wise Adaptive Pruning, a training-free method for batched inference in Large Reasoning Models (LRMs), achieving a 39.7 percentage point average accuracy improvement over prior adaptive pruning a…

Researchers introduced BCIJelly, a unified Python ecosystem integrating 18 curated brain-computer interface (BCI) datasets, 15 benchmark decoders, and 80 reusable modules, plus an automated architecture search (AAS) and …

In 1997, Sepp Hochreiter and Jürgen Schmidhuber published "Long Short-Term Memory" in Neural Computation, introducing the LSTM architecture that solved the vanishing gradient problem in recurrent neural networks by using…

A 2014 study published in Frontiers in Human Neuroscience found that the experience of mathematical beauty activates the same brain reward regions, specifically the medial orbitofrontal cortex, as the appreciation of art…

A developer explains the mental model connecting neural networks, deep learning, Transformers, and attention to understand how LLMs work. The post traces the evolution from basic neural networks to the Transformer archit…

A researcher replaced the standard scaled dot-product attention in vision transformers with SSOG (Separable Sum of Gaussians), a content-blind geometric attention mechanism that uses a few Gaussian atoms per head and a c…

A new experiment called Wedge 1, run by an independent researcher, tested whether neural networks that 'grok' a task have found the true underlying rule or just a local approximation, using the Lennard-Jones potential as…

Machine learning is a sub-field of artificial intelligence that uses data to train predictive models, following a two-phase flow of computationally expensive offline training and fast, cheap inference in production. The …

WIRED's guide to expanding the mind examines how brains, from the multibrained octopus to brain organoids, function as inference engines making flawed bets on the future, presenting five insights for survival. The Aug 14…

Researchers introduced Householder Graph Neural Network (HouseGNN), a deep GNN architecture that counters oversmoothing by using Householder reflections and GroupSort to preserve Euclidean norms at every node and depth, …

A new arXiv preprint (2608.12447v1) reports that in all eighteen transformer models tested (dense and mixture-of-experts, 7B-120B, base and instruction-tuned), residual-stream variation is geometrically and behaviorally …

A new arXiv paper (2608.12597v1) introduces Random Mapping Networks (RaMaN), a method that predicts the minimum latent dimension needed to train neural networks via random low-dimensional reparameterization, reducing opt…

A commentary on McCoy & Griffiths (2025) argues that their method for distilling Bayesian priors into artificial neural networks via Model-Agnostic Meta-Learning (MAML) does not actually instill a prior, but merely initi…

Pathway researchers released BDH (Dragon Hatchling), a biologically inspired large language model architecture that matches GPT-2-scale Transformers on language and translation tasks at 10M–1B parameters while achieving …

Batch Normalization, a technique introduced to stabilize deep neural network training, has become a cornerstone of modern deep learning, particularly in convolutional neural networks for computer vision. The method norma…