Seventeenth-December 2015 will be another contest algorithms and machine vision systems - ImageNet Large Scale Visual Recognition Challenge. This event is held annually , starting in 2010 and has 50 members.
As reports EE Times, the first data published by Microsoft and Google. According to the source, the first machine intelligence ahead of human image recognition. Thus, the level of errors in a certain way was 5.1%. The error rate of hardware and software Microsoft was 4.94%. Algorithms Google were smarter Microsoft's complex : 0.04%. Sponsor of the event, by the way, is NVIDIA company , which provides participants with access to the remote array accelerators based on proprietary GPU.
All competitors configure hardware computing resources in the form of neural networks at different levels. This approach proved to be the most promising at this stage, although the method of deep learning systems, each company had its own But basically, it all comes down to one approach, where the training system used a huge database of images.