Pupils Fast.Ai, a small organization conducted by free online courses on machine learning, were able to create an artificial intelligence algorithm, which, according to the test, exceeds Google’s specialist code.
The result of the group shows that not only those who have enormous resources can be engaged in the successful development of projects in the field of AI. Written algorithm based on Dawnbench test from Stanford researchers turned out to be more efficient than Google. This test uses the overall task of classifying images to track the speed of the code in the calculation of one dollar computing power.
AI students studied in just 18 minutes using Amazon Web Services, and about $ 40 was spent on the calculation. The head of online courses argues that this indicator is about 40% lower than that of the Internet giant, although it is difficult to compare due to various equipment.
Despite the fact that Google conducted tests on their own chips designed specifically for machine learning, and Fast.Ai disciples used AMAZON cloud services available, they were able to show the best results. This has already noted the representative