Sign in

GSoC started four months ago and it is not just about knowing more about the open-source that made the experience great! My mentors made it way cooler than I thought it would be. I was writing my Master thesis, for the last three months and surely, it has been a super productive summer for me! The best part is I get to do things at my own pace. My project particularly hasn’t been very easy to implement. I need to bridge a Machine Learning algorithm in the existing codebase. The fun part is venturing with different notebooks and figuring out…

In the last blog, I wrote about Introduction to JAX and Automatic Differentiation. In this one, my plan for the next stage of implementation. Currently, I am working on the modeling notebook ( to re-design it using JAX, especially to make optimization more robust by having JAX compute gradients on the likelihood function.

My mentor Daniela highlighted the issue that the current implementation is not robust using NumPy. The plan is to keep working on the current modeling notebook replacing NumPy by jax.numpy and also use grad, jit, vmap, random functionality of JAX.
When it comes to re-design, understanding the current…

I assume everyone reading this is already aware of two classical forms of differentiation, namely symbolic and finite differentiation. Symbolic differentiation operates on expanded mathematical expressions which lead to inefficient code and introduction of truncation error while finite differentiation deals with round-off errors. Optimized calculation of derivatives is crucial when it comes to training neural networks or mathematical modeling using bayesian inference. Both classical methods are slow at computing partial derivatives of a function with respect to many inputs, as is needed for gradient-based optimization algorithms. Here, automatic differentiation comes to the rescue. …

Raj Rashmi

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store