Ruaridh is a student of physics, living and studying in Scotland with the exception of this year where he’s in Canada on exchange.
As much as he likes physics, he also enjoys reading around the field, with a view to someday start something new. Of particular interest is intelligence, how the brain achieves it and how machines might be able to. Also, with a long-term interest in startups and business, he is currently working on (among other things) creating social networks based around common interests.
Besides his main interests, he likes exploring interesting ideas, finding new places, cooking tasty food, and generally enjoying life.
Project: Optical Character Recognition Using Deep Learning Techniques
My project is to try out optical character recognition using a new type of machine-learning technique: deep learning. This involves stacking copies of a machine-learning algorithm on top of each other, with the ouput from one going into the input of the next. I will be following George Hinton’s approach and using restricted Boltzman machines as the machine-learning algorithm.
First I will need to build a working implementation of a Boltzman machine and learn how to train it, using the guide published by Hinton. Then my task will be to try stacking two machines and looking at their performance and also at how they learn.
During this process I will be creating visualizations of the various aspects of the algorithms and getting an idea of how they learn.
Favorite Four-Color, Four State Turing Machine