Research map / Architectures and efficiency / RNNs and normalization
Tangent: Automatic differentiation using source-code transformation for dynamically typed array programming
Bart van Merrienboer, Dan Moldovan, Alexander Wiltschko
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Tangent's source-code-transformed derivative instead of PyTorch's | Scalar function with a tight loop · Tight scalar loop | runtime speed improvement: approximately 30% faster | Improved |
| Lazily initialize gradients with ZeroGrad rather than allocating zero arrays | Tangent · Array gradient accumulation | runtime overhead | Improved |
| Apply static dead-code elimination and algebraic simplification to generated adjoints | Tangent · Generated adjoints | unnecessary computation | Improved |
Closest papers by meaning
- Automatic differentiation in ML: Where we are and where we should be going (NeurIPS 2018 · similarity 0.89)
- Automatic Differentiation in Machine Learning: a Survey (JMLR 2018 · similarity 0.85)
- Automatic Differentiation of Programs with Discrete Randomness (NeurIPS 2022 · similarity 0.85)
- Automatic Functional Differentiation in JAX (ICLR 2024 · similarity 0.83)
- Collapsing Taylor Mode Automatic Differentiation (NeurIPS 2025 · similarity 0.81)
- $\bf{\Phi}_\textrm{Flow}$: Differentiable Simulations for PyTorch, TensorFlow and Jax (ICML 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
