Google-owned DeepMind has applied reinforced learning techniques to the multiplication of mathematical matrices, beating some human-made algorithms that have lasted 50 years and working toward improvements in computer science.
Founded in London in 2010, DeepMind has become famous for beating the world champion at board game Go with its AlphaGo AI and taking on the mind-bogglingly complex challenge of protein folding with AlphaFold.
In a wheels-within-wheels move, it has since set its sights on mathematical problems themselves.
Specifically, the lab said it developed a way to automate the discovery of algorithms that act as shortcuts when multiplying matrices – the cause of headaches for many a teenage math student.
For years, mathematicians have been applying algorithms to perform these complex array multiplications, some of which are used in computer science, particularly in machine learning and AI.
We’re told that DeepMind researcher Alhussein Fawzi and his colleagues used deep reinforcement to rediscover earlier matrix multiplication algorithms and find new ones. The team created a system, dubbed AlphaTensor, that plays a game in which the goal is to find the best approach to multiplying two matrices. If the AI agent does well, it is reinforced to make future success more likely.
This process is repeated over and over using this feedback so that agent generates interesting and improved ways to multiply matrices. It’s said that DeepMind’s agent was challenged to complete matrix math work in as few steps as possible, and had to figure out the best way forward from potentially trillions of possible moves.
We note that this AI agent was likely using matrix math in its learning process and during inference; thus, matrix operations were used to find faster ways to do matrix operations.
Fawzi told a press briefing this week the work was complex though resulted in the development of algorithms for matrix operations that have not been improved on in more than 50 years of human research, he said.
The researchers claimed the techniques could benefit computational tasks that use matrix multiplication algorithms – such as AI – as well as demonstrate how reinforcement learning can be used to find new and unexpected solutions to known problems, while also noting some limitations.
Skeptics may point to the application of AlphaFold, which promised breakthroughs in drug discovery via AI-supported protein research. Although the model has predicted nearly all known protein structures discovered, its ability to help scientists discover new drugs remains unproven.
In any case, this to us looks like machine learning being used to accelerate machine learning. Check out the above-linked research and paper for the full gory details on how it all worked.
Source: theregister.com
