| Popis: |
Deep learning uses multiple layers of artificial neurons for classification and pattern recognition. The biggest drawbacks of deep learning algorithms have been the high computation cost, inter-processor communication bottlenecks and parameters training time. Hence, incorporating parallel computing into deep learning decreases the computation time of complex deep learning algorithms. This chapter presents how parallelization is applied over many processors which are loosely coupled. Up to 4096 processes are scaled linearly with higher accuracy and zero loss percentage. This capacity of huge scaling helps in training billions of training examples in just a few hours. Various applications of Hessian-free parallelization mechanism on bioinformatics applications are in gene therapy, drug development, antibiotic resistance research, waste cleanup, climate change studies, bioweapon creation, improving nutritional quality and veterinary science. |
