AlphaDev: Discovering Faster Sorting Algorithms with Reinforcement Learning

Go back to programme

AlphaDev: Discovering Faster Sorting Algorithms with Reinforcement Learning

  • Watch

    * Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay

  • We engage with core computer algorithms such as “sorting” for arranging data on an enormous scale, exceeding trillions of daily instances. As our society becomes increasingly digitized, leading to an elevated demand for computational prowess, it is imperative that these algorithms function rapidly and seamlessly. While we have achieved substantial advancements in enhancing their speed in the past, refining these algorithms for heightened efficiency has posed a challenging task for both humans and computers alike. 

     

    AlphaDev represents an advancement of AlphaZero, an AI system that has demonstrated superior proficiency in games like Go, Chess, and Shogi. However, AlphaDev engages in a unique single-player game, one aimed at uncovering the most rapid sorting algorithm possible. In this quest, AlphaDev has succeeded in revealing several compact sorting algorithms that surpass those designed by humans. These discoveries have been assimilated over the previous year into one of the quintessential computer libraries, facilitating its daily use by millions of enterprises, developers, and users globally. 

     

    AlphaDev’s capacity for generalization extends beyond merely sorting algorithms. It has also discovered an accelerated hashing algorithm, another vital computer process, which has been similarly integrated into a core computer library. We project that our sorting and hashing algorithms combined are invoked trillions of times daily, underlying their critical importance in our increasingly digital world. 

    Share this session
    • Registration

      Watch

      * Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay

    • Start date
      20 July 2023 at 17:00 CEST Geneva | 08:00-09:30 PDT, San Francisco | 23:00-00:30 CST, Beijing
    • End date
      20 July 2023 at 18:30 CEST Geneva | 08:00-09:30 PDT, San Francisco | 23:00-00:30 CST, Beijing
    • Duration
      90 minutes (including networking)
    • Programme stream
    • UN SDGs