The question of artificial intelligence or AI's ability to create its own algorithms to speed up matrix multiplication has long been asked.
VentureBeat describes matrix multiplication as one of the most fundamental tasks of machine learning. In a recently published paper, DeepMind unveiled the first AI system called AlphTensor for discovering novel, efficient, and "provably correct algorithms."
Surpassing the #Strassenalgorithm has long been a #computerscience and mathematics holy grail. Today, @DeepMind unveiled #AlphaTensor, the “first #artificialintelligence system for discovering novel, efficient and provably correct algorithms.” https://t.co/A2exUXht8x
— VentureBeat (@VentureBeat) October 5, 2022
This Google-owned laboratory said the research sheds light on a five decades old open question in math about how to find the fastest way to multiply two matrices.
Since the 1969 publication of the Strassen algorithm, computer science has been in search to surpass its speed of multiplying two matrices.
Use of AlphaTensor AI System
While matrix multiplication is one of the simplest operations of algebra, which is taught in high school math, it is one of the most fundamental computational tasks as well, and as it turns out, it is one of the core mathematical operations in the present time's neural networks.
Essentially, matrix multiplication is used for processing images for smartphones, understanding speech commands, generating computer game graphics, compression of data, and more.
Today, companies are using costly GPU hardware to boost matrix multiplication efficiency so any additional speed would be game-changing when it comes to reducing costs and saving energy.
In a blog post on DeepMind's website, AplphaTensor is described as building upon AlphaZero, an agent that has shown superhuman "performance on board games like chess and Go."
This new tech project is taking the AlphaZero journey further, moving from playing games to dealing with unsolved mathematical problems.
AI Used to Boost Computer Science
The new study published in the Nature journal delves into how artificial intelligence could be used to boost computer science itself, explained Poshmeet Kohli, DeepMind's head of AI for science.
He added if they are able to employ AI to discover new algorithms for fundamental computational tasks, this has tremendous potential because they might be able to go beyond the algorithms presently used, which could lead to enhanced efficiency.
Kohli also explained that this is a particularly challenging task, since the process of discovering new algorithms is extremely difficult, and automating algorithmic discovery using AI needs a long and difficult reasoning process, from the formation of intuition about the algorithmic problem to writing an innovative algorithm and proving that it is correct on certain instances.
Artificial Intelligence for Both Science and Math
In July this year, researchers demonstrated that the AlphaFold tool of DeepMind could forecast the structures of over 200 million proteins from roughly a million species, covering almost every known protein on Earth.
According to Kohli, AlphaTensor shows the potential that AI not only in science but in mathematics, as well.
To find out if artificial intelligence is fulfilling that promise to go outside what human scientists have been able to do for the past five decades, "It is personally exciting," Kohli said.
He added that it is just showing the amount of effect that artificial intelligence and machine learning can have.
Related information about AlphaFold is shown on DeepMind's YouTube video below:
Check out more news and information on Artificial Intelligence in Science Times.
