DeepMind AI creates sorting algorithm 70% faster than humans, making software around the world faster

According to news on June 8, AlphaDev, an artificial intelligence developed by Google's DeepMind, created a new data sorting method that is 70% faster than algorithms designed by human programmers. Discovery has the potential to make millions of software around the world run faster.

Data sorting is one of the basic functions of computers, used to arrange data in alphabetical order or numerical size. There are already many different sorting algorithms, but due to decades of optimization, it is difficult for humans to innovate.

The design goal of AlphaDev is to discover new algorithms to complete a given task and try to surpass existing methods. Rather than fine-tuning current algorithms, it's built from scratch. It uses an intermediate language called assembly code, which is closer to the binary instructions of a computer than code written by humans, and makes it easier for AlphaDev to create more efficient algorithms.

AlphaDev generates one instruction at a time, and then tests whether its output is correct, while also requiring the generation of the shortest algorithm. DeepMind says the task is difficult because of the sheer number of possible instruction combinations.

When asked to create a sorting algorithm, AlphaDev surprisingly found a method for sorting five pieces of data that was 70% faster than the best algorithm. Even when sorting 250,000 data, it is 1.7% faster than the best algorithm.

"We initially thought it was wrong or had some bug, but when we analyzed the program, we realized that AlphaDev had actually found a faster way." said Daniel Mankowitz of DeepMind.

Since sorting algorithms are widely used in a variety of commonly used software, this improvement may have a significant impact on global computing. DeepMind has made them open source and added them to a common code base called Libc, which means anyone can use them immediately. This is the first time that the sorting algorithm part of this code base has been updated in more than a decade.

Mankowitz said that Moore's Law (IT House Note: The regular doubling of chip performance) is coming to an end due to physical limits, but AlphaDev may be able to make up for this by improving efficiency.

Mark Lee of the University of Birmingham in the UK said that AlphaDev is interesting and that even a 1.7% speed increase is useful. But he said he's skeptical that this approach can make up for the end of Moore's Law because it's unlikely to achieve the same progress in more complex software.

The above is the detailed content of DeepMind AI creates sorting algorithm 70% faster than humans, making software around the world faster. For more information, please follow other related articles on the PHP Chinese website!