Winner of the Pierre Robillard Award
Saifuddin Syed is the winner of the Pierre Robillard Award of the Statistical Society of Canada. Saif's thesis, entitled "Non-reversible parallel tempering on optimized paths", was written while he was a doctoral student at the University of British Columbia working under the supervision of Alexandre Bouchard-Côté.
Saif's research aims at developing robust and scalable algorithms for Bayesian statistics. The fundamental object in Bayesian statistics is the posterior distribution, which requires numerical techniques such as Markov Chain Monte Carlo (MCMC) to make inferences which can be unreliable when applied to the complex, high-dimensional posteriors commonly faced by practitioners. Parallel tempering (PT) is a popular technique that uses parallel computing to improve the reliability of MCMC algorithms. Traditionally, PT relied on the reversibility principle, which makes it challenging to tune and even deteriorating performance with too many parallel threads.
In his thesis, Saif identified a non-reversible PT variant (NRPT) that probably dominates its reversible counterpart. Moreover, he showed NRPT improves with additional parallelism, making it scalable to GPUs. Saif then designed a new theoretical framework to study NRPT and an efficient algorithm to tune it optimally. Scientists have since used NRPT at scale to tackle challenging Bayesian inference problems in diverse fields such as cancer genomics, nuclear physics, political science, and astronomy. Most notably, in 2021, the Event Horizon Telescope collaboration used NRPT to discover magnetic polarization in the photograph of supermassive blackhole M87 and to generate the first image of Sagittarius A*, the supermassive black hole at the center of the Milky Way.
Saif earned a BMath in pure and applied mathematics from the University of Waterloo and an MSc from the University of British Columbia in mathematics. He is now a Florence Nightengale Fellow in computational statistics and machine learning at the University of Oxford's Department of Statistics.
The criteria used in selecting the winner of the Pierre Robillard Award include the originality of ideas and techniques, the possible applications and their treatment, and the potential impact of the work. The award is named in memory of Professor Pierre Robillard, an outstanding dynamic young statistician at the Université de Montréal, whose untimely death in 1975 cut short what promised to be a highly distinguished career.
Saifuddin will present an overview of his work at a special session at 2023 SSC Annual Meeting.
The citation for the award reads:
To Saifuddin Syed, for the thesis entitled "Non-Reversible Parallel Tempering on Optimized Paths".