Machine Learning & AI Trick Derivations

This is the simplest way I can think of illustrating the probabilistic machine learning concept. Priors x Likelihoods give you posteriors over unknowns, i.e., parameters in our model. Given the posterior, now you can make a prediction via marginalisation. This, in turn, gives you uncertainty estimates which are like the big deal; telling us how sure we are about our predictions!

Notation based on those awesome slides:https://www.cs.toronto.edu/~radford/ftp/bayes-tut.pdf

Generative models are the big deal today with dalle 2 and stable diffusion on the fly. I was doing a course on score based generative models and I think this is one of the most influential results!

Text to image is awesome and a big deal! Stable diffusion is completely open source and you can use it!! Learn how here

Stable Diffusion: Tutorials, Resources, and Tools - Stack Diary This article covers introductory information on Stable Diffusion, as well as tools for generating art, and tutorials on how to use the AI model effectively.

We continue with the beauty of multi-variate Gaussians. Turns out, their squared Wasserstein between has a nice closed form as well! Just beautiful😃

KL-regularisation is everywhere in Machine Learning. Here's a step by step proof of the KL between two Gaussian distributions.

Amazing book!

We got lots of requests for hosting live sessions on the mathematics of machine learning. Is there any interest in joining those?

We are teaching a great course this year at the Oxford ML course. Come join me and lots of other fantastic speakers.

Haitham Bou Ammar on Twitter “I mean look at this amazing pannel of speakers. Apply! https://t.co/WudScbaXJb”

Policy gradients are a set of very successful algorithms in reinforcement learning, especially in robotics.

Have a look at this nice survey from world leaders in that field:

https://spiral.imperial.ac.uk/bitstream/10044/1/12051/7/fnt_corrected_2014-8-22.pdf

is full of non-convex optimisation problems. Here's a nice resource for non-convex optimisation in ML:https://arxiv.org/pdf/1712.07897.pdf

Numerical Optimisation is a critical ingredient in ML. It all starts in the convex world. Check out the go-to-book:https://web.stanford.edu/~boyd/cvxbook/bv_cvxbook.pdf