Marc Deisenroth Lecturer, Imperial College London |
|
How DID YOU GET INTO MAchine LEARNING?
I have a Computer Science MEng degree from the University of Karlsruhe. In the last two years of this degree, I focused on optimal control theory. Without any significant ML background I started a machine learning PhD with Carl Rasmussen at the Max Planck Institute for Biological Cybernetics in 2006. At that time, machine learning was not as hyped up as now, and I considered a PhD centred around Gaussian Processes and Reinforcement Learning an interesting and important extension of the control research that I conducted during my MEng. Soon, I realised that there is so much more to machine learning, an area that overlaps with statistics, computer science, engineering and social sciences. The wealth of opportunities offered by machine learning keeps me excited about the field to date. I am very grateful that I had the opportunity to move into the field without having a strong ML background.
WhAT WILL YOU Be teaching?
I will be teaching a short course on Mathematics for Machine Learning. The aims of this course is to set the main mathematical foundations for machine learning in general and deep learning in particular. I will be starting with a revision of calculus and matrices before moving to vector calculus and the chain rule. I will also review some principles of probability and briefly discuss some standard probability distributions.
What TOPICs Excite you most at the moment?
I am very excited about data-efficient machine learning. Data-efficient machine learning is concerned about extracting valuable information from data to facilitate fast learning in complex environments. Topics that fall into this category are probabilistic modelling, active learning, transfer learning, Bayesian optimization and model-based reinforcement learning.
