Google scientists create software that enables AI to develop without human input

Google scientists create software that enables AI to develop without human input

Google scientists are ready to develop a self-developed machine learning inspired by Darwinian evolution. Google is at the forefront of advanced artificial intelligence (AI) systems and tools. Late last year, Tech Titan created a machine learning model that enables developers to “investigate perspective model
behaviors” into machine learning models.

This revolutionary set of tools and frameworks, known as Explanable AI, will help developers and data scientists unravel the complexities of machine learning models and gain enhanced insights. Recently, the search giant has unveiled another breakthrough in machine learning. Theoretically, the basic premise of machine learning is a simple algorithm feeding a dataset, and the next time similar data is encountered it is ‘taught’ to respond in certain ways. But in practice, it is very difficult and that is why there is such a demand for experts like data scientists. Before anything close to a functional machine learning algorithm effective for executives, various steps need
to be taken, starting with collecting and preparing data, setting up the evaluation protocols, and developing benchmark models to create machine learning algorithms.

Google scientists create software that enables AI to develop without human input

Google scientists create software that enables AI to develop without human input

In short, machine learning is beyond the reach of many, and yet the rapid boom and endless applications mean more and more businesses now want to miss out, whether it’s to improve products and services for customers or to make internal processes more efficient. This enthusiasm has led many to consider ‘off-the-shelf’ machine learning solutions, and this is what machine learning automatically did to make ML-non-experts accessible to ML.

In a study published in the preprint server RXV, Google computer scientist Coke Lee and colleagues employed the concept of Darwinian evolution, such as natural selection, to enable ML algorithms to improve generations. Their program, a
combination of basic mathematical operations created 100 unique algorithms called AutoML-Zero that they then tested on common tasks such as test recognition.

After comparing their performance with hand-designed algorithms, the best ones were kept and small random “mutations” were introduced into their code, while the weaker candidates were removed. A highly-edited set of algorithms was
found as the cycle continued, some of which are comparable to classic machine learning techniques – such as neural networks (a type of computer program that loosely mimics how our brain cells work together to make decisions). This proves the team’s idea, Lee told Science Magazine, but he is hopeful that the processes will eventually be able to create more complex AI that human researchers have never been able to find. “Our goal is to make AutoML go even further: today it is possible to automatically discover complete machine learning algorithms using only basic mathematical operations as building blocks, the team wrote in a paper awaiting a paper-review.

“We have evolved linear registers, neural networks, gradient origins, multiplier interactions, weight averages, normalized gradients, etc., starting from empty component functions and using only basic mathematical operations. The author continues. These results are promising, but there is still a lot of work to be
done.”

Share This Post