NEUROEVOLUTIONARY ALGORITHMS FOR NEURAL NETWORKS GENERATING

Solving engineering problems using conventional neural networks requires long-term research on the choice of architecture and hyperparameters. A strong artificial intelligence would be devoid such shortcomings. Such is carried out a very wide range approaches: for example, biological (attempts to grow brain in laboratory conditions), hardware (creating processors) or software (using power ordinary CPUs GPUs). The goal work develop system that allow evolutionary approaches generate suitable solving problems. This called “neuroevolution”. purpose this also includes study features possible applicable strategies. object neuroevolutionary approach machine learning. subject strategies, coding methods organism’s genome. scientific novelty lies testing previously unused strategies generalization obtained systems “general intelligence”. simulating neuroevolution was created. specifics implementation were considered, algorithms justified, their explained. In order perform experiments, datasets created applying developed. It choose most optimal training parameters, find relationship between them, as well accuracy speed training. cannot said models implemented within directly bring us closer AI. They still lack own memory certain level complexity. For successful use, it necessary configure view input data some calculations outside model. However, future, can developed, with SNNs, use special equipment