2019/09/30 by Shin Kamada, Kamada, Shin, Takumi Ichimura +1
Computer Science · #Anomaly Detection Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #Digital Media Forensic Detection #FOS: Computer and information sciences #FOS: Electrical engineering #Generative Adversarial Networks and Image Synthesis #Image and Video Processing (eess.IV) #Neural and Evolutionary Computing (cs.NE) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1909.13480
openalex publication_date 2019/09/30 · openalex created_date 2020/09/25 · openalex updated_date 2026/07/28
Deep learning builds deep architectures such as multi-layered artificial\nneural networks to effectively represent multiple features of input patterns.\nThe adaptive structural learning method of Deep Belief Network (DBN) can\nrealize a high classification capability while searching the optimal network\nstructure during the training. The method can find the optimal number of hidden\nneurons of a Restricted Boltzmann Machine (RBM) by neuron\ngeneration-annihilation algorithm to train the given input data, and then it\ncan make a new layer in DBN by the layer generation algorithm to actualize a\ndeep data representation. Moreover, the learning algorithm of Adaptive RBM and\nAdaptive DBN was extended to the time-series analysis by using the idea of LSTM\n(Long Short Term Memory). In this paper, our proposed prediction method was\napplied to Moving MNIST, which is a benchmark data set for video recognition.\nWe challenge to reveal the power of our proposed method in the video\nrecognition research field, since video includes rich source of visual\ninformation. Compared with the LSTM model, our method showed higher prediction\nperformance (more than 90% predication accuracy for test data).\n