Date: Saturday, February 29, 2020 - 10:30
Venue: Martin Wood Lecture Theatre, Clarendon Laboratory
Prof Ard Louis
An Introduction to deep learning
Podcast Presentation (PDF)
In less than ten years, machine learning techniques based on deep neural networks have moved from relative obscurity to central stage in the AI industry. Large firms such as Google and Facebook are pouring billions into research and development of these new technologies. The use of deep learning in physics is also growing exponentially. I will give a basic introduction to deep learning for physicists and address a few questions such as: Is the hype around deep learning justified, or are we about to hit some fundamental limitations? Can physics help us understand why deep learning works so well? And conversely: How can deep learning provide new insight into the world around us?
Prof Andre Lukas
Machine Learning and String Theory
String theory produces the largest datasets currently known. Recently, string theorists have started to use methods from data science - particularly machine learning - to analyse the vast landscape of string data. I will discuss some of these developments and explain how they might help to answer questions in string theory.
Dr Elliott Bentine
Machine learning techniques in modern quantum-mechanics experiments
Modern table-top experiments can engineer physical systems that are deeply into the quantum mechanical regime. These cutting-edge instruments provide new insights into fundamental physics, and a pathway to future devices that will harness the power of quantum mechanics. They typically require complex operations to prepare and control the quantum state, involving time-dependent sequences of magnetic, electric and laser fields. This presents experimental physicists with an overwhelming number of tunable parameters, which may be subject to uncertainty or fluctuations. In this talk, we discuss how recent experiments have exploited machine-learning techniques, both to optimize the operation of these devices and to interperet the data they produce.