Ilias Diakonikolas

Digital Library

ACM Grace Murray Hopper Award

Greece - 2024

citation

For contributions to the field of algorithmic robust statistics by introducing new techniques to robustly estimate high-dimensional distributions along with a surprising variety of algorithmic applications

Diakonikolas' breakthrough work with collaborators beginning in 2016 gave the first efficient algorithms for learning the parameters of a high-dimensional distribution that are robust to arbitrary corruption in a constant fraction of data, with the constant independent of dimension. Subsequent work of Diakonikolas showed these new methods are not only of theoretical interest, but can be made practical, and can also be used to tackle more complex robust high-dimensional statistical problems, such as mixtures of distributions, in some cases being more efficient than even the prior state-of-the-art non-robust methods.

Diakonikolas' work broke barriers on the central question that stymied researchers since the 1960s in robust high-dimensional statistics. These problems have been widely studied in several communities, but were notoriously difficult and thought to be impossible to achieve with efficient algorithms. Indeed, Huber's book on robust statistics from nearly 30 years ago contained a remark essentially giving up on rigorous, provable methods to make progress on robust statistics for massive data, conjecturing that the community would have to settle for heuristics.  Diakonikolas' work provides new algorithmically efficient robust statistical estimators that perform well for real world data that may significantly deviate from idealized modeling assumptions.

Diakonikolas has also played a pivotal role in disseminating his new ideas, including by co-authoring a textbook (with D. Kane) that is bound to become the textbook of choice for courses on this topic.

Press Release

2024 ACM Grace Murray Hopper Award

Ilias Diakonikolas, a Professor at the University of Wisconsin, Madison, is the recipient of the 2024 ACM Grace Murray Hopper Award. He is cited for making contributions to the field of algorithmic robust statistics by introducing new techniques to robustly estimate high-dimensional distributions along with a surprising variety of algorithmic applications.

The primary focus of Diakonikolas' work is the mathematical foundations of data analysis, machine learning and algorithmic statistics. He is most well-known for his work on robust statistical algorithms for high-dimensional data. “Robust algorithms” are algorithms that perform well even when the data significantly deviates. The new paradigms that Diakonikolas developed changed the way we think about what is possible for efficient algorithms that process high-dimensional data—overcoming problems that have stymied researchers since the 1960s.

In 2016, Diakonikolas and collaborators gave the first efficient algorithms for learning the parameters of a high-dimensional distribution. These breakthrough algorithms are robust to arbitrary corruption in a constant fraction of data, with the constant being independent of the dimension. Subsequent work by Diakonikolas showed these new methods are not only of theoretical interest but can be made practical. In turn, these new methods can also be used to tackle more complex robust high-dimensional statistical problems. In mixtures of distributions, for example, there are cases where Diakonikolas’ algorithms are more efficient than even the prior state-of-the-art non-robust methods. Among other applications, robust algorithms are critical in order to develop reliable machine learning systems.

In addition to his breakthrough research contributions, Diakonikolas has also played a leading role in establishing a vibrant community around this work. He has organized workshops, given tutorials, and co-authored (with Daniel Kane) Algorithmic High-Dimensional Robust Statistics, a new textbook which has become an essential resource for this emerging field.