Deepmind introduces ‘AlphaTensor’, an Artificial Intelligence System (AI) for Discovering Novel, Efficient, And Exact Algorithms for Matrix Multiplication
It is important to improve the efficiency of fundamental computations, as this can have a major impact on the speed of many computations. Matrix multiplication is a simple task that can be found on systems such as neural networks and in scientific computing routines. Machine learning can go beyond the human intellect and even beat the best algorithms designed by humans. This process is complex due to the large number of algorithms that can be used. DeepMind made a major breakthrough by creating AplhaTensor. This is the first artificial intelligence system to develop new, efficient, and indisputable correct algorithms for basic operations such as matrix multiplication. They solved a 50-year-old mathematical puzzle: how to multiply two matrixes as fast as possible.
AlphaZero is the basis for AlphaTensor. This agent was a superhuman performer in board games such as chess and go. AlphaZero was able to progress from traditional board games, such as chess and go, to complex mathematical problems. This study, according to the team, is a significant milestone in DeepMind’s goal to improve science by using AI to solve fundamental problems. The research was also published in Nature, a prestigious journal.
The matrix multiplication algorithm is one of the simplest algorithms that students are taught in high school. However, it has many real-world applications. This method can be used for a variety of things including the processing of images on smartphones, identifying commands verbally, creating graphics for games, etc. Computing hardware that can multiply matrices efficiently consumes many resources. Therefore, even small improvements in matrix multiplication efficiency have a big impact. The study examines how AI can be used to improve the development of matrix multiplication algorithms. AlphaTensor relies on the human intuition to discover algorithms that are better than those currently available for a variety of matrix sizes. AlphaTensor’s AI-designed algorithm outperforms those designed by humans. This represents a significant advance in algorithmic discovery.