Unifying all Machine Learning Frameworks

In this talk, we will show how unifying all Machine Learning (ML) frameworks could save everybody a HUGE amount of time and energy. Through interactive coding sessions and live demos, we will explain how Ivy (checkout lets-unify.ai) is solving this unification problem. We will focus on demos using Ivy’s 3D vision and robotics libraries, solving 3D robotic navigation and perception tasks in a 3D simulator, all in real-time. Checkout https://github.com/ivy-dl/robot for examples! Finally, we will explore how you can join and contribute to the growing Ivy community, and help us in our mission to truly unify all ML frameworks once and for all.

Talk is based on the speaker's paper:
Ivy: Unified Machine Learning for Inter-Framework Portability
https://arxiv.org/abs/2102.02886
https://github.com/ivy-dl/robot

Lecture abstract

The number of open-source ML projects, libraries and codebases has grown considerably in recent years, and these are all written in a vast array of different incompatible ML frameworks. Wouldn’t it be nice if you could take the author's JAX code of an exciting paper and then immediately run it straight in your PyTorch pipeline without any issue? Ivy makes this possible. Ivy is a thin templated and purely functional framework, which wraps existing ML frameworks to provide consistent call signatures and syntax for the core tensor operations. Higher level functions, layers and libraries can then be built on top of Ivy’s functional API, for users of all frameworks. With the use of framework-specific frontends currently in development, Ivy will also enable automatic conversion between any two different frameworks. No need to “back a horse” with your framework selection, Ivy enables you to back all horses simultaneously, and mix and match libraries for all frameworks in a single project!

Presenter BIO

Daniel Lenton is currently undertaking his PhD in Robotics and 3D Vision under the supervision of Prof. Andrew Davison in the Dyson Robotics Lab, Imperial College London. He currently serves as a reviewer for NeurIPS, CVPR, IROS, ICRA and others. He is also CEO and Founder of Ivy, which has just raised a round of pre-seed VC funding to hire a team of developers. Ivy is on a mission to unify all Machine Learning (ML) frameworks. Daniel has also interned at Amazon Prime Air, working on real-time drone vision systems, and applying Generative Adversarial Networks (GANs) for dataset augmentation to train object detectors. Prior to his PhD, Daniel completed his MEng Mechanical Engineering also at Imperial College, attaining 1st class honors and deans list.
More information can be found at https://djl11.github.io/

-------------------------
Find us at:

All lectures are uploaded to our Youtube channel ➜ https://www.youtube.com/channel/UCHObHaxTXKFyI_EI8HiQ5xw

Newsletter for updates about more events ➜ http://eepurl.com/gJ1t-D

Sub-reddit for discussions ➜ https://www.reddit.com/r/2D3DAI/

Discord server for, well, discord ➜ https://discord.gg/MZuWSjF

Blog ➜ https://2d3d.ai