CS Department welcomes Ilias Diakonikolas, Assistant Professor in Algorithms and Machine Learning

Ilias Diakonikolas

Ilias DiakonikolasIlias Diakonikolas comes to the Department of Computer Sciences at UW-Madison from the University of Southern California and the University of Edinburgh. His research areas are algorithms and the foundations of machine learning, and some of his recently developed algorithms have been applied in discovering patterns in biological datasets and creating defenses against data poisoning attacks in machine learning. Diakonikolas disliked programming when he first started learning about computer science: “The reason was that I had not yet discovered the mathematical connections which made me fall in love with the field soon afterwards.”

Hometown: Athens, Greece

Educational/professional background:

I obtained my undergraduate degree in Greece, at the National Technical University of Athens. I obtained my PhD in Computer Science from Columbia University. After that, I spent two years at UC Berkeley as the Simons postdoctoral fellow in Theoretical Computer Science. Before joining UW, I held faculty positions at the University of Edinburgh and the University of Southern California.

How did you get into your field of research?

My undergraduate degree was in Electrical and Computer Engineering. While I enjoyed most of the topics I learned about during undergrad, I distinctly remember how enthusiastic I was about my first introduction to algorithms. Combined with a mathematical inclination, this naturally led me to theoretical computer science.

Could you please describe your area of focus?

The current focus of my research is on the algorithmic foundations of data analysis, with an explicit emphasis on designing efficient algorithms for fundamental problems in statistics and machine learning. My work is driven by two goals:

  1. Developing theoretically principled methods to solve fundamental machine learning problems; and
  2. Leveraging the insights underlying these methods to make machine learning systems more effective and reliable.

What main issue do you address or problem do you seek to solve in your work?

My research addresses important challenges in modern data analysis, including statistical and computational efficiency, robustness/model misspecification, privacy, and communication constraints. My contributions include efficient algorithms for a range of supervised and unsupervised estimation tasks in both low and high dimensional settings. In addition to their theoretical guarantees, several methods resulting from my work have been impactful in a number of pressing practical applications.

What attracted you to UW-Madison?

I was particularly attracted to the high research quality in the CS department, combined with the relaxed and collaborative culture of the faculty. A number of great colleagues within and outside CS convinced me that Madison is a great place to live and do research.

What was your first visit to campus like?

I first visited Madison for my job interview in Spring 2018. I distinctly remember the temperature difference between LA and Madison in March, which I’m still trying to get used to. More importantly, I was impressed by the in-person discussions with colleagues in the CS department. After the interview was over, I was convinced that Madison is a place where I would be happy to live and pursue my research agenda.

What’s one thing you hope students who take a class with you will come away with?

A major focus of my teaching effort is to communicate that the goal is understanding, rather than just following the material. In pursuit of this goal, my approach is to reward intellectual effort and discourage focus on grades.

Do you feel your work relates in any way to the Wisconsin Idea? If so, please describe how.

A major focus of my recent work has been to design principled methods that make machine learning systems more effective and reliable. A concrete example concerns designing effective defenses against data poisoning attacks, where a small fraction of fake data can substantially compromise the behavior of a machine learning system. Even though these are classical questions in statistics, until very recently, even the most basic computational questions in this field were poorly understood. My work has proposed a general methodology that led to highly scalable algorithms that provide the strongest known defenses against such attacks.

What course(s) are you teaching this year? Anything in particular you’re looking forward to regarding teaching?

In the Fall, I am teaching a new graduate class on algorithmic aspects of unsupervised learning. In the Spring, I will be teaching a new undergraduate class on the statistical and algorithmic foundations of supervised learning. It may sound cliché, but the best part about teaching for me is the interaction with the students. One of the most satisfying aspects of teaching for me is communicating my excitement about the material and inspiring students who want to transition to the research frontier.

What first interested you about CS? About your specific field within CS

Perhaps surprisingly, I had no experience in computer science before college. My parents got me my first desktop computer during the first semester of my undergraduate studies. In fact, in the beginning of my undergrad I was pretty sure I disliked programming! The reason was that I had not yet discovered the mathematical connections which made me fall in love with the field soon afterwards.

Where do you see your area of CS in the future – in 5, 10, or 25 years?

I try to refrain from making such predictions. The beauty of research is that we do not know what it will lead to and, unless we pursue it, we will not find out. Personally, I am very excited about the potential societal impact of algorithmic research in a variety of settings, including machine learning security, privacy, and fairness.