ABAB Pedia

Matej Balog

Overview

Contents9 sections
Key facts

Studies in Oxford, Cambridge and Tübingen

At Oxford, Balog earned a master’s degree in mathematics and computer science, working with Yee Whye Teh on applications of the Mondrian process. He joined the Cambridge–Tübingen machine learning doctoral program under Zoubin Ghahramani in Cambridge and Bernhard Schölkopf in Tübingen. During a year at the Max Planck Institute he investigated kernel methods in differential privacy with Ilya Tolstikhin and Schölkopf.⁠[3][1]

AI-assisted algorithm discovery

At Google DeepMind’s AI for Science unit, Balog works on automated algorithm discovery. His projects include AlphaTensor for matrix multiplication, FunSearch for mathematical and computational problems, and AlphaEvolve for searching for and improving algorithms.⁠[1][4]

Middle European Mathematical Olympiad and Oxford

In September 2010 Balog was a third-year student at Gymnázium Grösslingová in Bratislava. He represented Slovakia in the Middle European Mathematical Olympiad at Strečno, scoring 16 points for an individual bronze medal. His own record of his mathematics and computer science studies at Oxford begins with his first term in autumn 2011.⁠[5][6]

Program synthesis and doctoral research

In 2017 Balog and colleagues published DeepCoder, a paper investigating how machine learning could assist program synthesis. His 2020 doctoral dissertation in Cambridge’s Department of Engineering examined ways to convert computational tasks such as sampling into optimization problems, with applications that included differential privacy and program synthesis.⁠[2]

Interview with No Priors

Balog described sharing AlphaEvolve’s output with mathematicians who examined the code to understand how its steps constructed a solution.⁠[4]

Advice Shared on No Priors

He recommended framing problems so the result can explain a construction. The algorithm can matter more than an individual answer. (35:07)⁠[4]

He encouraged researchers to examine how a solution is built so that they can understand the underlying ideas. (35:41)⁠[4]

Related Organizations, Websites and Public Accounts

Google DeepMind: Research scientist. Official website: https://deepmind.google/ (opens in a new window).⁠[1]

Sources6 sources

  1. Matej Balog (opens in a new window)matejbalog.eu↑
  2. Matej Balog – MLG Cambridge (opens in a new window)mlg.eng.cam.ac.uk↑
  3. Matej Balog (opens in a new window)institute-tue.ellis.eu↑
  4. No Priors Ep. 120 | With Google DeepMind’s Pushmeet Kohli and Matej Balog (opens in a new window)youtube.com↑
  5. Mathematical Olympiad: MO results (opens in a new window)skmo.sk↑
  6. Oxford | Matej Balog (opens in a new window)matejbalog.eu↑