Pankaj Mehta uses statistical physics to study machine learning

January 30, 2015

How do computers learn to identify images or speech? Recent work by Assistant Professor Pankaj Mehta and collaborator David Schwab (Northwestern University) suggests that the answer lies in a statistical technique known as "renormalization". Their work shows that this technique, which allows physicists to extract relevant features from a particular system, is the same process used by artifical neural networks to categorize data. Read the full story in Quanta Magazine.