EPFL SIAM Student Chapter Seminar on AlphaTensor - Francisco Ruiz (Google DeepMind)


Event details

Date 05.12.2023
Hour 18:0019:00
Speaker Francisco Ruiz, Research Scientist @ Google DeepMind London (UK)
Location Online
Category Conferences - Seminars
Event Language English

The EPFL SIAM Student Chapter is honored to host Francisco Ruiz, Research Scientist @ Google DeepMind London (UK), who will give an online seminar on AlphaTensor. The event will be projected in MA B1 11.

Francisco Ruiz is a Research Scientist working at Google DeepMind in the Deep Learning Team. Before joining DeepMind, he was a Postdoctoral Research Scientist in the Department of Computer Science at Columbia University and in the Engineering Department at the University of Cambridge, where he held a Marie-Sklodowska Curie fellowship in the context of the E.U. Horizon 2020 program. He completed his Ph.D. and M.Sc. from the University Carlos III in Madrid. His research is focused on statistical machine learning, especially Bayesian modeling and inference.

Title of the Seminar: AlphaTensor: Discovering faster matrix multiplication algorithms with reinforcement learning

Abstract: Algorithms are the building blocks of computation, and their efficiency in solving fundamental computational tasks is crucial. Matrix multiplication is one such primitive task, occurring in many systems - from machine learning to scientific computing routines. Despite decades of research, the question of how efficiently we can multiply n x m by m x p matrices remains open. Here we develop a Deep Reinforcement Learning approach for discovering efficient and provably correct matrix multiplication algorithms. Our agent, AlphaTensor, is trained to play a single-player tensor decomposition game with a finite factor space, and yields algorithms for multiplying arbitrary matrices. AlphaTensor discovered new algorithms that outperform the state-of-the-art complexity for many matrix sizes (n, m, p). In particular, AlphaTensor improved on Strassen's two-level algorithm for 4 x 4 matrices in a finite field, which represents the first improvement since its discovery 50 years ago. We further demonstrate the flexibility of AlphaTensor through different use-cases: it discovered novel algorithms with state-of-the-art complexity for structured matrix multiplication, and improved the practical efficiency of matrix multiplication by optimizing for runtime on specific hardware. Our results highlight AlphaTensor's ability to accelerate the process of algorithmic discovery on a range of problems, and optimizing for different criteria.

Hope to see all of you there!

If you want to be informed about the latest event of the EPFL SIAM Student Chapter, please visit our website https://siam.epfl.ch or follow our LinkedIn page https://www.linkedin.com/company/epfl-siam-student-chapter/
If you want to join our association, drop a message to [email protected] !

Practical information

  • General public
  • Free
  • This event is internal


  • EPFL SIAM Student Chapter


  • Fabio Zoccolan, 


Applied Math Linear Algebra Reinforcement learning

Event broadcasted in