A consortium of 40 technology companies, including Facebook, Google and others, develop an assessment methodology and a set of criteria for testing artificial intelligence. By measuring AI products in these categories, companies will be able to determine the best solutions for them, learning technologies, and so on. The consortium itself was named MLPerf.
The benchmarks, called MLPerf Inference v0.5, focus on three common machine learning tasks: image recognition, object detection, and translation systems. Given the different processing capabilities of different devices, there are separate tests for AI on different platforms such as smartphones, servers, and chips.
MLPerf General Chairman Peter Mattson noted that the formation of such AI evaluation criteria will create standards for evaluating new machine learning software platforms, hardware accelerators, and cloud and edge computing platforms in real-life situations. It will also create a level playing field that even the smallest companies can use.
In addition, standardization will spur the development of AI systems and direct it in the right direction. At the moment, it is not reported what technical features are implemented in MLPerf Inference v0.5. It is also not clear when the test suite will hit the market and become available to commercial entities. However, the very fact of the appearance of such a tool indicates that developers see prospects for AI systems.
Source: 3dnews.ru