Representational image.  |  Photo Credit: iStock Images

DeepMind's AI is probably best known for cracking the popular strategy game Go, but in the last few years, machine learning has proved extremely valuable in an array of applications like protein-folding and deep intuition.

Now, for the first time, the technology has been used to identify mathematical connections that have eluded researchers for decades. Teaming up with mathematicians, DeepMind's AI sought to tackle two distinct problems one in the study of symmetries and the other in knot theory.

I was very struck at just how useful the machine-learning tools could be as a guide for intuition, said Marc Lackenby, one of the mathematicians from the University of Oxford who participated in the study. I was not expecting to have some of my preconceptions turned on their head.

The study of math may turn a lot of people off but, at its core, it facilitates a greater human understanding of the fundamental properties that govern our universe. It's only through painstaking work in the area of pure mathematics that we now have revolutionary technologies like airplanes and computers.

Mathematicians try to spot patterns in large datasets which they then seek to formulate conjectures out of. These conjectures are then reviewed and tested by their peers in various hypothetical cases and, if they hold up, turn into theorems.

But the amount of data now available is impossible for any human to process. And this is where machine learning comes in. Machine learning can discover patterns much quicker than humans insights that can then guide new mathematical ideas.

Take the theory of knots for example. At a superficial level, knots describe how a piece of string or rope is entangled. But at a much deeper level, they revolve around key mathematical principles that can be applied in the realm of quantum computing.

Algebra, geometry, and quantum theory all share unique perspectives on these objects and a long-standing mystery is how these different branches relate: for example, what does the geometry of the knot tell us about the algebra? wrote the researchers.

The team of researchers created a machine learning model to probe these connections and one particular trick called saliency maps proved immensely valuable. The ML model was able to spot specific geometrical properties known as a 'signature' that researchers could then use to formulate a conjecture.

In another instance, DeepMind teamed up with mathematicians to probe a problem in symmetries one that scientists have traditionally studied using charts or graphs. But as more data is incorporated, these charts inevitably grow dauntingly large, making it nearly impossible for a human to comprehend. But DeepMind's AI discovered numerous interesting patterns that, the researchers, think could guide mathematicians toward a proof.

I was just blown away by how powerful this stuff is, said Dr Geordie Williamson from the University of Sydney. I think I spent basically a year in the darkness just feeling the computer knew something that I didn't.

DeepMind has been consistently proving that the applications of machine learning extend well beyond just games, and the latest breakthrough is another testament to the technology's growing value in solving some of humanity's most complex problems. But ultimately, due to its inherent probabilistic nature, it needs to be accompanied by human intuition and rigour. Nevertheless, the man-machine combination, the researchers believe, could inspire other scientists to incorporate AI into their own research.

Here is the original post:
In breakthrough, DeepMind's AI has cracked two mathematical problems that have stumped experts for decades - Times Now