Thanks to a artificial intelligence developed by British company DeepMind, an algorithm widely used around the world can run up to 70% faster.
This improvement is due to a new approach found by Artificial Intelligence to classify data that had been neglected by human programmers for decades.
see more
Google Messaging: The future of cross-platform chat could be…
Confirmed: Samsung really is producing foldable screens for…
The discovery has the potential to optimize millions of programs, as sorting algorithms are fundamental functions widely used by computers around the world.
Daniel Mankowitz of DeepMind expressed surprise with the results obtained, stating that they did not expect superior performance.
He highlighted the conciseness of the developed program, which was studied for decades. Classification algorithms, widely used in computing, play a key role in the organization of data, whether through alphabetical ordering of words or classification numeric.
Although there are several approaches in this field, innovations have been limited due to the high optimization achieved over time.
Sorting algorithms, known as the “workhorses” of computing, are tasked with organizing data, whether sorting words alphabetically or sorting numbers in ascending order.
Despite the existence of several classification algorithms, innovations have been limited, as these approaches have been highly optimized over the decades.
To overcome this limitation, DeepMind developed an Artificial Intelligence model called AlphaDev. As opposed to tweaking existing algorithms, AlphaDev is designed to discover new algorithms capable of performing a given task.
This innovative approach starts from scratch, seeking to overcome previous efforts in the field of data classification. DeepMind uses assembly code as part of the AlphaZero development process.
Assembly code is an intermediate computer language, lying between code written by humans and sequences of binary instructions encoded in 0s and 1s.
While assembly code can be carefully read and understood by humans, most software is written into high-level languages, which are more intuitive, before being translated or “compiled” into code assembly.
However, DeepMind argues that the use of assembly code gives AlphaGo greater freedom to create more efficient algorithms by taking advantage of their ability to understand the language in depth level.
During the development process, the artificial intelligence is guided to build an algorithm step by step and test its output against a known correct solution, thus ensuring the creation of a method effective. In addition, the AI is instructed to seek to create the most concise algorithm possible.
The company points out that this task becomes increasingly challenging as the problems become more complex. This is because the number of possible combinations of instructions increases rapidly, approaching the number of particles in the universe.
Thus, AI has to face the difficulty of exploring an extremely vast search space to find optimal solutions to larger problems.
AlphaDev, when challenged to create a classification algorithm, surprised by presenting a result 70% faster than the algorithm considered the best for lists of five pieces of data. Furthermore, in lists with more than 250,000 items, the algorithm developed by AlphaDev achieved a performance improvement of 1.7%.
Daniel Mankowitz, when analyzing the program created by AlphaDev, initially suspected an error or flaw, however, when investigating further, realized that the AI had discovered a more efficient approach to classifying data.
Lover of movies and series and everything that involves cinema. An active curious on the networks, always connected to information about the web.