To comprehend complex, interdisciplinary engineering problems with broad applicability, Guy Bresler develops mathematical models.
Guy Bresler’s research may potentially be represented visually using something akin to a Venn diagram. He is employed at the four-way intersection of theoretical computer science, probability, statistics, and information theory.
“New tasks are constantly needed at the interface. According to Bresler, an associate professor in the Department of Electrical Engineering and Computer Science at MIT who recently received tenure, “there are always prospects for completely new issues to explore (EECS).
As a theoretician, he seeks to comprehend the complicated interactions between data structure, model complexity, and the amount of computation required to learn those models. Finding the “sweet spot” where adequate data and computer resources allow academics to efficiently solve an issue has been his main focus in recent times. These fundamental phenomena are generally responsible for determining the computational complexity of statistical problems.
The struggle between data and computation is common when attempting to tackle a challenging statistics problem. Without enough data, a statistical problem’s required computation may prove impossible to perform or at the very least require a surprising amount of resources. However, given just the right amount of information, the seemingly insoluble can be resolved, and the amount of processing required to do so is drastically reduced.
This type of trade-off between compute and data is present in the bulk of contemporary statistical issues, with applications ranging from drug discovery to weather forecasting. Cryo-electron microscopy is a further well-researched and practically significant example, according to Bresler. With this method, scientists acquire pictures of molecules in various orientations using an electron microscope. The main difficulty is how to resolve the inverse problem, which involves figuring out the structure of the molecule from noisy data. This type of inverse problem can be used to formulate many statistical issues.
Clarifying connections between the numerous distinct statistical topics currently under study is one goal of Bresler’s work. As has been done for other kinds of computer issues in the subject of computational complexity, the goal is to categorise statistical problems into equivalence classes. He claims that by demonstrating these linkages, academics can apply their understanding of one problem to another rather than attempting to comprehend each issue separately.
Using a theoretical strategy
Bresler’s decision to enter academia was motivated by his desire to theoretically comprehend different basic occurrences.
He claims that his parents, who were also academics, demonstrated how rewarding a career in academia can be. His father, an electrical engineer and theoretician researching signal processing, provided him with his first exposure to the theoretical side of engineering. His work served as an early source of inspiration for Bresler. At the University of Illinois at Urbana-Champaign, he took physics, math, and computer science courses in various combinations. But regardless of the subject, he tended to favour the theoretical point of view.
Bresler relished the chance to work on a wide range of subjects while a graduate student at the University of California, Berkeley, including probability, theoretical computer science, and mathematics. His passion for learning new things served as his primary motivator.
There is a feeling that one had best learn as much as possible if one is to have any chance of discovering the correct instruments to answer those issues when working at the intersection of various fields with new problems, he says.
His interest in information and decision systems (LIDS) at MIT attracted him there in 2013 for a postdoc. Two years later, he joined the faculty as an assistant professor in EECS. In 2019, he received the title of associate professor.
Bresler claims he was lured to MIT’s intellectual milieu and the climate that encouraged taking on risky research endeavours and striving to advance in new fields of study.
Possibilities for cooperation
“What really surprised me was how alive, dynamic, and cooperative MIT is. I’ve got a mental list of more than 20 folks that I’d want to have lunch with once a week and work on research projects with. So joining MIT was definitely a victory solely on the basis of numbers,” he says.
Working with his pupils has been a particular highlight for him because they are constantly teaching him new things and posing insightful queries that inspire fascinating study ideas. One of these pupils, Matthew Brennan, who was close to Bresler, died tragically and unexpectedly in January 2021.
Bresler is still reeling from Brennan’s death, which temporarily put a stop to his studies.
He had the incredible talent to listen to a nearly entirely incorrect concept I had, pull a valuable piece from it, and then pass the ball back. “Beyond his own prodigious powers and ingenuity,” he says. “We were driven to attempt to tell a certain tale, and we shared the same vision for what we wanted to accomplish in the job. It was in a manner lonely at the time because so few people were working in this particular field. But because of our mutual trust, even when things were hopeless, we encouraged one another to press on.
These lessons in tenacity serve as fuel for Bresler as he and his students continue delving into questions that are challenging to resolve.
He has spent more than ten years working intermittently in the area of learning graphical models from data. He notes that domain specialists who have the necessary knowledge and skills frequently create models of specific types of data, such as time-series data made up of temperature readings.
However, it is not at all clear what shape a model should take for many types of data with complex connections, such as social network or biological data. Bresler’s research aims to estimate a structured model from data, which might subsequently be applied for later applications like better weather prediction or providing suggestions.
The fundamental problem of finding appropriate models, whether analytically or algorithmically in a complex environment or by defining a helpful toy model for theoretical analysis, he claims, ties the theoretical work with engineering practise.
Modeling is generally a form of art. Real life is complex, thus any model that attempts to capture every aspect of a situation through writing would fail, according to Bresler. To identify the proper elements of the problem to be modelled and have any chance of truly solving it and learning what one should do in practise, one must think about the issue and have some understanding of the practical side of things.
Bresler frequently finds himself solving issues that are considerably different from those in the lab. He loves to boulder all over New England and spends a lot of his leisure time doing so.
“I adore it so much. It serves as a solid justification for venturing outside and entering a whole new environment. Even if there is some problem solving involved and philosophical parallels, it is completely unrelated to sitting down and doing math, the author claims.