Deep Learning on multi-GPU

Deep Learning becomes more and more important nowadays. It is necessary to put mass data into deep learning model and repeatedly practice the model to gain more useful features,and thus a strong computing power is essential. With the advent of Graphics Processing Unit (GPU), CPU cannot meet the requirements any more. GPU can provide ten and even hundreds times performance than CPU and multi-GPU devotes to effectively execute multiple concurrent threads on multi-cores, which is certainly a best computing power for deep learning. Image recognition is an important direction in Deep Learning, so it becomes our aim in this dissertation.