Efficient Automatic Optimization of Neural Network Architecture

Abstract

Neural Networks are at the heart of deep Learning Frame works which have yielded excellent results in various complex problem domains. But the design of neural network architecture is a challenging task. Judicious selection of network architecture and manual tuning of network parameters is a tedious and time consuming process. There has been a substantial e ort to automate the process of neural network design using various heuristic algorithms. Evolutionary algorithm are amongst the most successful methods to automate the network architecture search process. But these algorithms are very computation intensive. Thus we try to explore a technique that could lead to faster evolutionary algorithms to nd optimal neural network architecture.We also do a survey of various alternative methods.