Fast and scalable neuroevolution deep learning architecture search for multivariate anomaly detection

The neuroevolution is one of the methodologies that can be used for learning optimal architecture during the training. It uses evolutionary algorithms to generate topology of artificial neural networks (ANN) and its parameters. In this work, a modified neuroevolution technique is presented which incorporates multi-level optimization. The presented approach adapts evolution strategies for evolving ensemble model based on bagging technique, using genetic operators for optimizing single anomaly detection models, reducing the training dataset to speedup the search process and performs non gradient fine tuning. The multivariate anomaly detection as an unsupervised learning task is the case study on which presented approach is tested. Single model optimization is based on mutation, crossover operators and focuses on finding optimal window sizes, the number of layers, layer depths, hyperparameters etc. to boost the anomaly detection scores of new and already known models. The proposed framework and its protocol shows that it is possible to find architecture in a reasonable time which can boost all well known multivariate anomaly detection deep learning architectures. The work concentrates on improvements to multi-level neuroevolution approach for anomaly detection. The main modifications are in the methods of mixing groups and single models evolution, non gradient fine tuning and voting mechanism. The presented framework can be used as an efficient learning network architecture method for any different unsupervised task where autoencoder architectures can be used. The tests were run on SWAT and WADI datasets and presented approach evolved architectures that achieve best scores among other deep learning models.