Researchers from London have implemented a reinforcement-learning-based AI system to improve math-based algorithms
Computer programming implements of maths for analyzing and then manipulating representations of real-world phenomena. The most important math calculation involved in programming is matrices. Carrying out matrix calculations, requires a lot of work, especially as the matrices grow larger. Hence, a team of researchers at Google’s DeepMind, London, has demonstrated that AI can find faster algorithms to solve matrix multiplication problems.
The researchers at DeepMind imagined the possibility of using a reinforcement-learning-based AI system to generate new algorithms with fewer steps than those implemented now. They found that most of the gaming systems implemented reinforcement learning. The team shifted its focus to tree searching after completing preliminary systems, which are also used in game programming. When applied to multiplying matrices, the researchers found that converting an AI system to a game allowed for searching for the most efficient way to get to the desired outcome—a mathematical result.
The researchers experimented with their system by enabling it to find, review, and implement existing algorithms, employing rewards to attract them to pick the most efficient one. The system learned about the factors that enhance matrix multiplication efficiency. The researchers also forced the system to generate its algorithms for again improving efficiency. In many instances, this experiment demonstrated that the system’s algorithm was more efficient than those created by human predecessors.
Click here for the Published Research Paper