The research and development partnership between the major web companies like Google and the Chinese search engine Baidu, and the research centers specialized in Harvard and California University, has resulted in this scientific result concerns the data scientists who currently have a tool performance measurement, which also makes it possible, according to logical evolution, to improve the analyzes. Technically, MLPerf makes it possible to calculate the speed by giving the importance to the time necessary for the realization of stains with different dimensions as with example the knowledge of the object and the possibility of measuring its recognition in relation to the intended object.
With the demand for more and more speed on artificial intelligence and machine learning projects, MLPerf is a benchmark that has several functions:
DeepSpeech2 (Librispeech) Translation – Transformer (WMT English-German) Sense Analysis-Seq-CNN (IMDB dataset) Reinforcement learning – Mini-go (prediction of movement in professional games).
NG Perf, one of MLPerf’s promoters, said: “AI is changing a lot of industries, but to realize its greatest potential, we need faster hardware and software.’ Also, Steve Conway, senior vice president of research at Hyperion Research, said: “MLPerf is a useful and excellent tool because in recent years there has been a lack of real standards for buyers and sellers. This benchmark seems to be specific to early problems in artificial intelligence. Most of them are there is a boundary problem. Later we needed other criteria because artificial intelligence began to have borderless issues. Problems with borders are relatively simple, such as speech, image recognition, or games. Unbounded line issues include cancer diagnosis and reading magnetic resonance report now; artificial intelligence can propose recommendations for complex problems. ”
Initially, the engineers of this project chose to release the first version to measure the level of acceleration in a market. It mainly relates to the performance of training tasks in workplaces, particularly on information systems dedicated to data processing in the data center. Similarly, to make this tool an indicator of the measures, part of the set will be devoted to predictive analysis and decision support to improve the customer relationship.
