Discovering faster matrix multiplication algorithms with reinforcement learning

TLDR

Improving the efficiency of fundamental algorithms, such as matrix multiplication used in neural networks and scientific computing, can have widespread impact, but automating algorithm discovery is challenging due to the enormous search space. The study aims to use deep reinforcement learning, specifically AlphaZero, to automatically discover efficient and provably correct matrix‑multiplication algorithms beyond current human‑designed ones. AlphaTensor trains a single‑player game to find tensor decompositions within a finite factor space, employing a deep reinforcement learning framework based on AlphaZero. AlphaTensor discovered matrix‑multiplication algorithms that beat the state‑of‑the‑art complexity for many sizes, including a new 4×4 algorithm that surpasses Strassen’s two‑level method, and demonstrated flexibility by producing efficient algorithms for structured matrices and hardware‑optimized implementations, thereby.

Abstract

Abstract Improving the efficiency of algorithms for fundamental computations can have a widespread impact, as it can affect the overall speed of a large amount of computations. Matrix multiplication is one such primitive task, occurring in many systems—from neural networks to scientific computing routines. The automatic discovery of algorithms using machine learning offers the prospect of reaching beyond human intuition and outperforming the current best human-designed algorithms. However, automating the algorithm discovery procedure is intricate, as the space of possible algorithms is enormous. Here we report a deep reinforcement learning approach based on AlphaZero 1 for discovering efficient and provably correct algorithms for the multiplication of arbitrary matrices. Our agent, AlphaTensor, is trained to play a single-player game where the objective is finding tensor decompositions within a finite factor space. AlphaTensor discovered algorithms that outperform the state-of-the-art complexity for many matrix sizes. Particularly relevant is the case of 4 × 4 matrices in a finite field, where AlphaTensor’s algorithm improves on Strassen’s two-level algorithm for the first time, to our knowledge, since its discovery 50 years ago 2 . We further showcase the flexibility of AlphaTensor through different use-cases: algorithms with state-of-the-art complexity for structured matrix multiplication and improved practical efficiency by optimizing matrix multiplication for runtime on specific hardware. Our results highlight AlphaTensor’s ability to accelerate the process of algorithmic discovery on a range of problems, and to optimize for different criteria.