Neural-network building blocks · MLD-0077
Automatic differentiation
Computing derivatives by applying differentiation rules to executed elementary operations.
Original reference page · detailed expansion pending. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.
Original example
Obtain parameter gradients from a tensor program.
Original caveat
Different from symbolic differentiation and finite differences.
Context and aliases
Level: Intermediate. Stages: Training · Inference · Theory.
Tags: DL · SLM · LLM · Vision · NLP.
Aliases: Autodiff · AD.
Related concepts
Concepts linking here
Technology connections
- Burn — Rust-native deep-learning framework.
- Equinox — JAX library representing models as PyTrees.
- Flax — Neural-network library built on JAX.
- JAX — Composable automatic differentiation, vectorization and compilation for numerical Python.
- Keras — High-level model-building API supporting multiple numerical backends.
- MLX — Array and ML framework optimized for Apple silicon.
- PyTorch — Tensor computation, automatic differentiation and neural-network framework.
- TensorFlow — Tensor and neural-network framework with deployment tooling.
- tinygrad — Compact tensor and automatic-differentiation framework.
Original references
- Deep Learning: online textbook ↗ · Consulted
- PyTorch neural-network layers ↗ · Reference link
- PyTorch documentation ↗ · Reference link
- Original reference URL ↗