Event
Getting Started with Slang: Automatic Differentiation
Automatic Differentiation
Join us for a focused deep dive on automatic differentiation in Slang with Shannon Woods. This session distills the core concepts of autodiff, emphasizing how forward and reverse modes work, how gradients propagate, and how to reason about correctness and efficiency with clear, minimal examples.
What we’ll cover:
- Forward and backward gradient propagation: intuition and step‑by‑step examples
- How to invoke autodiff from Slang
- Diff pairs and the forward/backward operators
- Differentiable vs. non‑differentiable types; making custom structs differentiable
- Providing custom derivatives for fine‑grained control
- Handling buffer access and gradient accumulation patterns
- Practical techniques for debugging and validating gradients
Who should attend:
- Graphics engineers exploring optimization or inverse problems
- Researchers and practitioners in differentiable rendering or neural graphics
- Developers curious about practical, GPU‑friendly autodiff concepts
- Bring your questions—there will be time for discussion and hands‑on gradient debugging tips.