Neural Networks

Deep neural networks face training challenges like vanishing gradients and slow convergence. Techniques such as activation functions, Xavier and He weight initialization, and Batch Normalization help overcome these issue…

Ganesh, building the AI code reviewer *git-lrc*, explains how to calculate gradients for hidden layer weights and biases in backpropagation. The derivation uses the chain rule to propagate error from the output layer to …

A study in Cell reveals that bilingual brains use a shared neural map to represent word meanings across languages, based on recordings from individual brain cells in the hippocampus. The findings suggest the brain stores…

A recent engineering graduate built a millimeter-wave radar prototype that uses FMCW technology and a neural network to classify building materials, specifically targeting asbestos detection. The project, which took six …

Researchers at Aix-Marseille University and the Collège de France discovered a neural bridge between fear and physical reactions by monitoring rat hippocampus activity during fear conditioning. The study, published in PN…

Scientists at Shimane University reversed autism-like symptoms in mice by repairing shortened axon initial segments in neurons of the medial prefrontal cortex. The study, published in Cell Death & Disease, showed that ac…

Chebyshev polynomials, named after Pafnuty Chebyshev, are sequences of orthogonal polynomials defined by trigonometric identities. They are used in polynomial interpolation and approximation due to their extremal propert…

Meta AI released Brain2Qwerty v2, a non-invasive brain-to-text pipeline that decodes typed sentences from magnetoencephalography (MEG) signals with 61% word accuracy, up from 8% for prior methods. The system uses a convo…

Meta unveiled Brain2Qwerty v2, a non-invasive brain-to-text decoder that achieves 61% average word accuracy using MEG recordings, approaching invasive BCI performance. The system, trained on 22,000 sentences from nine vo…

Researchers at arXiv propose that meta-learning, or learning to learn, may explain why human visual representations are more flexible than those of pretrained neural networks. By training a sequence model on thousands of…

Researchers propose DCSNet, an end-to-end framework for small object segmentation in medical imaging that integrates Detection-guided Hierarchical Cropping and Multiscale Feature Aggregation to overcome class imbalance a…

Researchers introduced AEGIS, a semantic-aware adversarial detection framework for vision sensors, combining a SemantiGAN module and evidential deep learning to filter adversarial inputs and provide calibrated uncertaint…

Researchers introduced turn-averaged sparse autoencoders (SAEs) that represent entire conversational turns with fixed feature sets, improving interpretability of long-context language models. The method outperforms per-t…

Researchers prove that mesh intelligence requires a liquid substrate—a continuous-time, multi-timescale network—because agents must estimate a changing latent from irregular, unscheduled observations without retraining. …

Researchers introduced the Covariance-Agnostic Neural Kalman Consensus Filter (CA-NKCF), a distributed sensing framework that combines partial domain knowledge with deep neural networks for latent state estimation withou…

Researchers propose S-GAI, a spectral geometry-aware initialization framework for sigmoidal MLPs that encodes dataset geometry into network weights. Using SVD to extract class-wise spectral features, S-GAI initializes hi…

Researchers used polynomial algebra to study critical points in deep monomial networks, finding that for large activation degrees, criticality occurs at subnetworks where neurons are inactive or redundant. This provides …

Researchers propose a gravitational interpretation of fine-tuning reversion, where large early training phases create dominant behavioral manifolds that later fine-tuning reverts toward. They identify a reversion directi…

Researchers propose a memory-augmented LSTM autoencoder for unsupervised human activity recognition using IMU sensor fusion, achieving 96.6% and 98.4% accuracy on DaLiAc and PAMAP2 datasets, outperforming supervised base…

Researchers submitted a system to the SoccerNet 2026 Player-Centric Ball Action Spotting challenge that uses a two-stage pipeline with per-player attention and agreement-based ensembling, improving Macro-F1 from 48.6 to …