Here's the next article in my series on ODEs and SDEs in ML for RBC Borealis. We show how to solve SDEs by changing the variable using Itô’s famous lemma, which we derive. We find closed-form solutions for geometric Brownian motion and the Ornstein-Uhlenbeck process. rbcborealis.com/research-blo...

Here is part VI of my series of tutorials on ODEs and SDEs in machine learning: rbcborealis.com/research-blo... To solve SDEs we must integrate the noise term and to this end, we develop the stochastic integral. It's solution is a stochastic process with mean zero and a time-varying variance.

Here is the 4th instalment in my series on ODEs and SDEs in machine learning. I previously discussed closed-form solutions for ODEs, but often there is no known solution. This article considers numerical methods, which can approximate the solution of any ODE. rbcborealis.com/research-blo...

Wow. Understanding Deep Learning has now been downloaded half a million times. Thank you so much everyone! I was overjoyed when it hit 100k so this is completely mindblowing. I'm so thrilled that people are finding it useful.

Exciting news! @travislacroix.bsky.social (who co-wrote the chapter on ethics in Understand Deep Learning) has a new book out "AI and Value Alignment". Recommended for anyone serious about ethics and AI. Details at: value-alignment.github.io Buy it here: broadviewpress.com/product/arti...

Here is part III of my series for @RBCBorealis on ODEs and SDEs in machine learning. This article develops methods for solving first-order ODEs in closed form; we divide ODEs into different families and develop approaches to solve each family. rbcborealis.com/research-blo...

Here's the 2nd part of my series on ODEs and SDEs in ML. This article introduces ODEs and is suitable for novices: rbcborealis.com/research-blo... We describe ODEs, vector ODEs and PDEs and categorize ODEs by how their solutions are related. We describe conditions for an ODE to have a solution.

I'm starting a series of articles on ODEs and SDEs in ML for RBCBorealis. I'll describe ODEs and SDEs from first principles without assuming prior knowledge and present applications including neural ODEs, and diffusion models. Part I: rbcborealis.com/research-blo.... Follow for parts II & III.

These blogs for RBC Borealis consider infinite-width neural networks from 4 viewpoints. We use gradient descent or a Bayesian approach, and, for each, we focus on either the weights or output function. This leads to the Neural Tangent Kernel, Bayesian NNs and NNGPs. Enjoy! tinyurl.com/yfsts565

Learning or teaching from my book (udlbook.com)? I have now added the complete bibfile (which is accurate and took ages to make) and the LaTeX for all of the equations (helpful if you are making slides).

Boris Meinardus: How I'd learn ML in 2025 (if I could start over) www.youtube.com/watch?v=_xIw.... (me too 😄)

Tutorial 4 of 4 on Bayesian methods in ML for RBC Borealis concerns Neural Network Gaussian Processes: rbcborealis.com/research-blo... Think your network might perform better if you increased the width? NNGPs are networks with INFINITE width! Includes code and links to background info on GPs.

Blog 3 of 4 on Bayesian methods in ML for RBC Borealis concerns Bayesian Neural Networks (i.e., Bayesian methods for NNs from a parameter-space perspective): rbcborealis.com/research-blo... Parts 1 and 2 (linked in article) introduced Bayesian methods. Coming soon in part 4: NNGPs