Abstract:We establish a fuzzy deep model called the fuzzy deep belief net (FDBN) based on fuzzy restricted Boltzmann machines (FRBMs) due to their excellent generative and discriminative properties. The learning procedure of an FDBN is divided into a pretraining phase and a subsequent fine-tuning phase. In the pretraining phase, a group of FRBMs is trained in a greedy layerwise way: the first FRBM is trained by original samples, and the average values of the left and right probabilities produced by its hidden units are treated as the training data for subsequent FRBMs. The resulting FDBN is either a generative or a discriminative model depending on the choice of training a generative or a discriminative type of FRBM on top. Then, a hybrid learning approach is proposed to fine-tune this novel fuzzy deep model: the well pretrained fuzzy parameters are first defuzzified, and the FDBN with defuzzified parameters is fine-tuned by the wakesleep or stochastic gradient descent algorithm. This hybrid strategy not only avoids learning an intractable fuzzy neural network, but also greatly improves the classification capability of the FDBN. The experimental results on MNIST, NORB, and 15 Scene databases indicate that the FDBN with the hybrid learning approach can handle high-dimensional raw images directly. It inherits the fine nature of the FRBM and outperforms some state-of-the-art discriminative models in classification accuracy. Moreover, it shows better capability of robustness than a deep belief net when encountering noisy data.

DBN classification algorithm for numerical attribute

Hyperbolic Binary Neural Network

Classification of Lung Nodules Based on Convolutional Deep Belief Network

An Improved DBN Method for Text Classification

Face recognition algorithm based on Gabor feature and DBN

Improved Classification with Semi-supervised Deep Belief Network

A DBN for Hyperspectral Remote Sensing Image Classification

Classification of Drug and Non-Drug Using Deep Belief Network Based on Data Preprocess

3D Object Classification Using Deep Belief Networks.

Data Classification Based on Density Function Estimated by Radial Basis Function

AN Information Text Classification Algorithm Based on DBN

Study on metric decomposition for DBN structure learning

A Novel Deep Belief Network Based on Shallow Feature Regression

A Fuzzy Deep Model Based on Fuzzy Restricted Boltzmann Machines for High-dimensional Data Classification

A Classified Identification Deep-Belief Network for Predicting Electric-Power Load

DBNet: A New Generalized Structure Efficient for Classification

Deep Belief Network for Spectral–Spatial Classification of Hyperspectral Remote Sensor Data

Classification of hyperspectral image based on deep belief networks

Machine Fault Classification Using Deep Belief Network.

Discriminative deep belief networks for visual data classification

Application of Deep Belief Networks for natural language understanding