Alan Milligan receives Best Master’s Thesis Award from national AI association
MSc graduate Alan Milligan honoured by CAIAC for outstanding work on machine learning optimization problems
Alan Milligan, a recent MSc graduate of UBC Computer Science, has been awarded the Best Master’s Thesis Award by Canadian Artificial Intelligence Association / Association pour l'Intelligence Artificielle au Canada (CAIAC). The award recognizes remarkable Master’s theses in the field of AI from students at a Canadian university.
“It’s a great honour to receive the award,” says Alan Milligan. “While it’s nice to get an award, I must credit much of the work to the many collaborators and mentors I’ve had at UBC. I hope that their contributions will also be further shown with this recognition.”
Alan’s thesis focused on the Adam algorithm that is used to train many machine learning models, including AI chatbots such as ChatGPT.
Similarly to how a guitar needs to be tuned in order to be in the right keys, machine learning problems require tuning different settings, or optimizing several parameters, for the model to produce the correct output. Adam is known to adapt to some part of the optimization problem for better performance, but researchers don’t have a clear idea of what Adam is adapting to.
Alan’s thesis explores how different parts of machine learning optimization problems can make the problems more difficult and how this affects Adam’s performance. He found that the structure in different parts of the optimization problem creates problems that are difficult for classical algorithms — but Adam is able to exploit these structures and remain effective.
“The projects Alan worked on have provided simple explanations to some well-studied and challenging topics related to the dominant algorithm for training machine learning models,” says Professor Mark Schmidt, who supervised Alan during his MSc. “His works are likely to influence the way we think about training machine learning models for years to come."
Prior to joining the Master’s degree program in computer science, Alan completed his BSc in Combined Honours Computer science and Mathematics at UBC. Now, as a PhD student at Mila – Quebec Artificial Intelligence Institute and Université de Montréal, he continues his study of machine learning algorithms.
“I hope to use the tools I’ve gained in my MSc to study problems in reinforcement learning, which is becoming more and more important in training the systems that impact our daily lives,” Alan says. “I’m interested in trying to give the machine learning community an easier time solving difficult optimization problems, whether that be through algorithms that are easier to run or require less tuning, or giving practitioners better intuitions for what makes their problem hard.”
He attributes his success to having great collaborators and coauthors who worked with him on his project, including his advisors and colleagues at UBC Computer Science.
“My thesis would not have been possible without the help of many people, and I largely have them to thank for it and the rest of my degree.”