Neural Networks news and analysis on Web Pulse: 1261 curated articles tracking the latest Neural Networks developments, tools, and research, updated continuously from vetted sources.
A new study on arXiv (2607.16231v1) evaluates OrthoGrad, a geometric intervention that removes the radial component of weight gradients during optimizer updates, in noisy-label image classification. On MNIST with small-d…
Researchers introduce the Diffusion-corrected Auto-Regressive Fourier Neural Operator (DiffARFNO), a two-stage framework combining an autoregressive Fourier-MIONet with a conditional Denoising Diffusion Implicit Model (D…
Researchers propose Group-Contrastive Forward-Forward (GCFF), a biologically plausible learning algorithm that yields monosemantic neurons organized into hierarchies of increasing abstraction without sparsity constraints…
Researchers developed a lightweight 1D CNN for affective touch classification in soft plush companions, achieving 75% test accuracy and 85% leave-one-subject-out cross-validation accuracy with a 13.2k-parameter model. Th…
Researchers propose Supermartingale-based Label Transition (SLT), an anchor-free loss correction framework for robust quantum neural network (QNN)-based medical image classification under noisy labels. SLT models entropy…
Researchers from MIT and other institutions have developed iTracker, a convolutional neural network for eye tracking that works on commodity mobile phones and tablets without additional sensors, achieving a prediction er…
An artificial neuron is a simple math function with four components: inputs, weights, a bias, and an activation function. Weights are initialized to small random numbers to break symmetry, and the bias shifts the neuron'…
Liquid Neural Networks (LNNs) use continuous-time dynamics and input-dependent time constants to adapt to streaming physical data, unlike Large Language Models (LLMs) which process discrete tokens. Despite common misconc…
Asankhaya Sharma released Neural Drive, a 130-million-parameter model that generates a Mario Kart-like racing game in a browser without a game engine or any code, imitating Super Tux Kart.
A study comparing deep learning architectures for sleep apnea detection from EEG signals achieved a best test AUC of 0.750 using a Vision Transformer trained on topological data analysis features, tested on 575 pediatric…
A new study from arXiv:2607.15449v1 applies renormalization group theory to Transformer attention, finding that attention's relevance depends on data correlation length. For long-correlation sequences, attention is a rel…
A new study comparing handcrafted feature descriptors and convolutional neural networks (CNNs) for facial expression recognition finds that CNNs achieve the best overall performance across three datasets (FER-2013, CK+, …
ImprovedVBGS, an accelerated framework for on-the-fly continual reconstruction, reduces mean per-frame latency from ~84.0 s to ~0.050 s on an RTX 3070 Ti using the NeRF synthetic dataset, a 1680x speed-up while maintaini…
A new blog post argues that existing methods for decoding internal representations in large language models (LLMs), such as Logit Lens, Tuned Lens, Patchscopes, and Anthropic's Jacobian Lens, are all heuristic and lack a…
Veso R&D released 'Knowledge as Dynamics', a research paper demonstrating that a neural network's latent vector field can be copied, merged, and inherited without training data or teacher weights, achieving 84% basin agr…
A new interpretability analysis of the 124-million-parameter GPT-2 checkpoint (revision 607a30d) demonstrates a three-layer evidence framework—pattern evidence, logit attribution, and intervention—to establish causal mec…
A new explainer drawing illustrates the geometric principles behind diffusion models, flow matching, and related generative techniques, tracing the lineage from denoising diffusion probabilistic models (DDPMs) through la…
A new analysis of brain imaging data published in the European Journal of Neuroscience reveals that LSD reduces the synchronization of local brain activity, particularly in visual and sensory regions, to produce its mind…
A developer on Hacker News has released a forward-only, autograd-free Physics-Informed Neural Network (PINN) that uses zero-copy memory interlock to reduce VRAM consumption to approximately 1/1000 of legacy backprop chai…
Sakana AI's Error Diffusion method trains dual-stream excitatory/inhibitory networks that obey Dale's principle without backpropagation, achieving 96.7% accuracy on MNIST and 61.7% on CIFAR-10. The approach sidesteps the…