google/neural-tangents

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

2383 stars · 238 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Neural Tangents is a Python library built on JAX for defining, training, and evaluating neural networks of both finite and infinite width. It computes the exact Gaussian Process kernels (NTK and NNGP) of complex architectures and enables analytic Bayesian inference or gradient-descent training in the infinite-width limit.

Use cases

• compute the neural tangent kernel of a CNN or transformer
• train an infinite-width network as a Gaussian process
• study gradient descent training dynamics of wide networks
• run exact Bayesian inference with NNGP kernels
• compare finite and infinite network behavior in research
• scale kernel computations across GPUs or TPUs

When to choose

• you need exact NTK/NNGP kernel computations for research
• you want to study infinite-width limits of neural networks
• you work in JAX and need composable layer primitives with kernel tracking
• you need multi-device distributed kernel computation

When to avoid

• you need a general-purpose deep learning framework for production models
• you require PyTorch or TensorFlow ecosystems
• your use case is standard finite-width training without kernel analysis

Facets

library · maturity stable

machine-learning deep-learning math sdk machine-learning deep-learning artificial-intelligence python cross-platform cli jax neural-tangent-kernel gaussian-processes infinite-width-networks bayesian-inference kernel-methods research gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google/neural-tangents")