Listen to this story
Back in school, we all solved matrix multiplications. The equations are used in every sphere of life, from processing images on smartphones to generating graphics for computer games. As more computational processes get involved, the maths gets complex. Researchers at Google’s DeepMind in London have introduced a new artificial intelligence (AI) system called AlphaTensor that can find shortcuts in this fundamental type of mathematical calculation.
The new AI system can discover efficient and correct algorithms for tasks such as matrix multiplication. The AI can turn the problem into a game by leveraging the ML techniques such as AIs used to beat human players in games such as Go and chess.
The AI system finds the fastest way to multiply two matrices, a question that has remained open for about 50 years. The paper was published in Nature, where the researchers said that “improving the efficiency of algorithms for fundamental computations can have a widespread impact, as it can affect the overall speed of a large number of computations”.
DeepMind’s approach uses a form of machine learning called reinforcement learning, in which an AI ‘agent’ (a neural network) learns to interact with its environment to achieve a multistep goal. If it does well, the agent is reinforced — its internal parameters are updated to make future success more likely.
DeepMind, in a statement, said, “AlphaTensor discovered algorithms that are more efficient than the state of the art for many matrix sizes. Our AI-designed algorithms outperform human-designed ones, which is a major step forward in the field of algorithmic discovery.”
The researchers are hopeful that the AI-enabled algorithms could make computational technology much more efficient, spurring new applications for designing algorithms that optimise metrics. They said, “We hope that our paper will inspire others in using AI to guide algorithmic discovery for other fundamental computational tasks.”