Neural Networks

Researchers at the University of Tennessee and the University of Florida developed a hardware neural network backdoor that can be hidden in custom chips used for edge AI, exploiting vulnerabilities in the supply chain of…

Noise Contrastive Estimation (NCE) and InfoNCE are machine learning methods for estimating statistical distributions by distinguishing real data from noise, with applications in language modeling, speech recognition, and…

Researchers at arXiv have causally demonstrated that the weight norm sets the grokking timescale in neural networks, settling a dispute over whether weight norm causes the delayed generalization. By intervening on the no…

Researchers introduced neural slack variables, a primal-side approach that converts constraint enforcement into a regression problem by coupling a primary network with an auxiliary network. The method achieved zero measu…

Researchers developed a Transformer-based scheduling policy for the open shop scheduling problem (OSSP) using deep reinforcement learning. The model, trained on small Taillard benchmark instances, generalizes to larger p…

Researchers propose Feature Interaction NEtwork (FINE), a lightweight semantic alignment module that refines low-level features via high-level contextual guidance using cross-level attention prior to fusion in object det…

A new thesis investigates how Bayesian principles can improve understanding of modern deep learning systems, introducing the Deep Variational Implicit Process (DVIP) and post-hoc methods VaLLA and FMGP for uncertainty es…

Researchers found that repetition loops in Gemma 4 instruction-tuned LLMs can be fixed by editing as few as one neuron, but the fix does not resolve deeper 'doom loops' caused by missing knowledge. The study shows that w…

Researchers introduced Muon$^p$, a new optimizer that uses fractional spectral-power updates to interpolate between Muon and gradient descent, improving finetuning performance on billion-scale models. The method preserve…

Researchers propose Spiking Fourier Graph Operators (SpikF-GO), a spiking neural network model for multivariate time series forecasting that uses a hypervariate graph to model inter-variable dependencies. SpikF-GO achiev…

Researchers developed the LANTERN framework, a neural network that models health-state transition probabilities for long-term care using irregular longitudinal data. The model outperformed logistic regression and gradien…

Researchers have demonstrated that convolutional neural networks can achieve higher accuracy by introducing learnable pairwise connections between filters, moving beyond traditional pointwise activation functions. The ne…

Researchers found that introducing structured noise into artificial neural network activations improves robustness against adversarial attacks and naturalistic image modifications, with noise structure from adversarial a…

A research project using k-sparse autoencoders found interpretable features in a small reasoning model, including features that appear to correspond to the model's reasoning process. The experiment analyzed hidden states…

The Kimi Team published a new attention architecture, Kimi Linear, on arXiv in 2025. The architecture uses Kimi Delta Attention interleaved 3:1 with full attention, becoming the first linear hybrid to outperform softmax …

Researchers Tian et al. introduced Visual Autoregressive Modeling (VAR), a new image generation method that predicts the next resolution instead of the next token, achieving state-of-the-art results on ImageNet and outpe…

Researchers at Beijing Normal University found that brain wave patterns measured at age nine can predict anxiety or depression in teenage years, based on a seven-year study of 64 children published in Biological Psychiat…

A developer has created Hillock, a brain-inspired memory system for local LLMs that runs entirely on CPU without requiring a neural vector database or VRAM. The system uses Hyperdimensional Computing (HDC) to store and r…

Researchers used language models and fMRI to show that neural representations of meaning are shared across English, Chinese, and French speakers, and that language models trained on different languages converge on a simi…

CoreAI released HTCE-Origin v1.0, a Q256 integer-only toroidal cognitive runtime designed for bounded, evidence-gated reasoning and adaptive memory. The system operates as a single runtime object with L1/L2/L3 state tran…