Representation of a convolutional neuronal network as a high level paralel composition applied to the recognition of DNA sequences

This work proposes the use of Structured Parallel Programming using the process communication pattern called Pipeline in its version of High-Level Parallel Composition (HLPC) to implement a process composition that represents a convolutional neuronal network or CNN and that is used to solve a specific problem of DNA sequences. The HLPC Pipeline-CNN is then shown, which represents the implementation of a convolutional neural network making use of the three types of parallel objects that make up an HLPC: A manager object, one or more stage objects and a collector object. The manager object represents the HLPC itself and makes an encapsulated abstraction out of it that hides the internal structure, the stage objects are objects of a specific purpose, in charge of encapsulating an client-server type interface that settles down between the manager and the slave-objects and the collector object that is an object in charge of storing the results received from the stage objects to which is connected. To show the usefulness and performance of the HLPC Pipeline-CNN implemented, it was used in the recognition of DNA sequences from a database with 4 types of hepatitis C virus (type 1, 2, 3 and 6). The results of this classification were obtained in terms of percentages of training precision and validation precision, as well as performance results in terms of speedup from 1000 to 4000 training steps with 2, 4, 8, 16 and 32 exclusive processors in one parallel machine of up to 64 processors with shared-distributed memory.