Enabling Fast Differentially Private SGD via Just-in-Time Compilation and Vectorization. Pranav Subramani, Nicholas…

arxiv.orgsite

Uses JAX's JIT and VMAP to achieve faster differentially private than existing libraries.

from

JAX

added

2026-10-10

likes

0

similar

1. ▲

   JaxDF github.com

   Framework for differentiable simulators with arbitrary discretizations.

   tool link ↗
   
2. ▲

   JAX github.com

   Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more.

   tool topic programming link ↗
   
3. ▲

   JAXopt github.com

   Hardware accelerated (GPU/TPU), batchable and differentiable optimizers in JAX.

   tool link ↗
   
4. ▲

   GPJax github.com

   Gaussian processes in JAX.

   tool link ↗
   
5. ▲

   Achieving 4000x Speedups with PureJaxRL chrislu.page

   A blog post on how JAX can massively speedup RL training through vectorisation.

   site link ↗
   site
6. ▲

   JAX, M.D. github.com

   Accelerated, differential molecular dynamics.

   tool link ↗
   

JAX › Papers: “Uses JAX's JIT and VMAP to achieve faster differentially private than existing libraries.”