Neural Networks

A developer shipped a browser version of a four-in-a-row board game where the opponent is a neural network, running entirely as static files with no backend. The network, a policy+value model with under a million paramet…

ChronoWeave is a system that uses large language models and neural networks to extract causal links from historical text and map them spatially, treating node positions as trainable tensors optimized via backpropagation.…

ChronoWeave is a system that extracts causal graphs from historical text and lays them out as interactive maps, using a neural network to position causally-connected events close together. The project's spatial intellige…

Hierarchical Navigable Small World (HNSW) is an approximate nearest neighbor algorithm that combines skip lists and navigable small world graphs to achieve logarithmic time complexity for similarity search in vector data…

A practical tutorial on LSTM interpretability demonstrates how to preprocess time-series data into a 3D tensor, build a stacked LSTM model with dropout and early stopping, and apply permutation importance to identify whi…

Neuromorphic computing offers a radical architectural departure from traditional AI by using local evolution, extreme sparsity, and event-driven processing to overcome the efficiency wall of current LLMs, according to a …

Principles of Intelligence (PrincInt, formerly PIBBSS) is launching PIRAMID, an internal research division that uses statistical physics to build scientific foundations for ambitious mechanistic interpretability. The div…

A Hacker News user argues that neuromorphic computing, which mimics the brain's local evolution, extreme sparsity, and event-driven processing, could provide radical efficiency improvements needed to overcome the scaling…

A developer published a guide covering the essential mathematical concepts that power modern AI and large language models, explaining how text is tokenized, embedded, and processed through a transformer pipeline to predi…

A developer demonstrates how to train an LSTM on daily temperature data and apply three explainability methods—permutation importance, SHAP, and Integrated Gradients—to interpret the model's predictions. The tutorial cov…

Researchers at TARA propose SONI (Selective Orthogonalisation via Noise Injection), a fine-tuning technique that uses targeted noise injection to selectively orthogonalize safety-critical feature directions in neural net…

A developer released Jixp, a Lisp DSL for describing Jax neural nets, as a side project for learning Jax. The compiler reads Racket-style DSL and outputs Python, enabling simple model specification with hidden implementa…

Researchers at the Universities of Konstanz and Stuttgart have built a computer from a few hundred microscopic particles oscillating in a liquid, using their collective dynamics for reservoir computing. Led by Clemens Be…

Researchers Timo Bolkart, Daoye Wang, and Prashanth Chandran introduced SHELLS, a method for topologically consistent multi-view 3D head reconstruction using a DinoV2 backbone with LoRA adaptation and an XCiT-based trans…

A new study from researchers at arXiv finds that the Muon optimizer's speedup in reaching the grokking threshold on modular arithmetic comes from orthogonalization via the Newton-Schulz iteration, not spectral-norm const…

Researchers present ODeform, a method extending Neural Ordinary Differential Equations to model continuous 4D dynamics of deformable objects in 3D space, eliminating discrete time steps while maintaining computational ef…

Researchers have identified new polynomial-time tractable neural network architectures for training with ReLU and linear activation functions, pushing beyond previous state-of-the-art results. For ReLU networks, the team…

Researchers have developed a zero-shot Digital Twin framework that uses a Thermodynamics-Informed Graph Neural Network to simulate physical systems on unseen geometries without retraining, operating at approximately 25 m…

Researchers introduced HypNO, a graph-based neural operator for scalar hyperbolic conservation laws that uses physics-informed message passing to respect upwinding and entropy admissibility near shocks. The architecture,…

Researchers propose Graph Wavelet Compressed Sensing (GWCS), a learning-based framework for offline compression of graph signals that combines a nonparametric multilevel importance sampler with a scale-aware graph neural…