[DEMO] High-Performance Quanvolutional Neural Networks with JAX & Flax

[Spartoons]

General information

Name
Aran Oliveras (@Spartoons)

Affiliation (optional)
Computer Vision Center (CVC), QML Group, Universitat Autònoma de Barcelona (UAB)

Twitter (optional)
@aranoliveras

Image (optional)

---

Demo information

Title
High-Performance Quanvolutional Neural Networks with JAX & Flax

Abstract
This demo presents a high-performance port of the classic Quanvolutional Neural Network tutorial, migrated from TensorFlow to the JAX/Flax ecosystem. By leveraging JAX's vmap and JIT compilation, it demonstrates how to vectorize the execution of quantum circuits over image patches, achieving massive speedups (over 10,000x faster processing per image compared to standard loops). It includes a custom, stateless training loop using Optax and Flax.

Relevant links

• Demo Repository: https://github.com/Spartoons/quanvolution-jax
• Original Paper: Henderson et al. (2019) "Quanvolutional Neural Networks"
• Original Tutorial: PennyLane Quanvolution

[CatalinaAlbornoz]

Thank you @Spartoons ! Do you also have a LinkedIn handle where we could tag you when doing marketing there?

[Spartoons]

Hi @CatalinaAlbornoz , thank you!

Yes, absolutely. You can tag me on LinkedIn here: https://www.linkedin.com/in/aran-oliveras/

Looking forward to it!

[CatalinaAlbornoz]

Thanks! We will get back to you with comments and next steps within a few weeks.

[CatalinaAlbornoz]

Hi @Spartoons, Happy New Year!

I wanted to let you know that we'll be ready to promote your demo in about 2 weeks! We'll add the links to the posts here so that you can reshare/repost. 🎉