Chenglong Zhang

This book is the culmination of our research in the recent decade on randomized neural networks with data-dependent supervision mechanisms. Traditional randomized neural networks mainly focused on constructing various deep neural networks with data independent random weights, ignoring the impact of the number of nodes and scope of parameters on the universal approximation property (UAP) of randomized neural networks. Comprising of 15 chapters, Advanced Randomized Neural Networks for Pattern Analysis introduces systematic solutions for advanced data-dependent stochastic configuration networks, namely algorithms that assign random parameters and construct network structures incrementally. The book is segmented into three major sections — neural networks optimization, robust data analysis, and deep fusion learning — that feature the successful performance of advanced randomized neural networks in various pattern analysis problems. We anticipate that both researchers and engineers in the field of artificial neural networks, particularly pattern recognition and medical diagnosis, will find this book and the associated algorithms useful, and we hope that anyone with an interest in the related research field will find the book enjoyable and informative.

Contents:

• Introduction
• Neural Networks Optimization:
  • Decay Regularized Stochastic Configuration Network with Multi-Level Signal Processing
  • Regularized Stochastic Configuration Network Based on Weighted Mean of Vectors
  • Stochastic Configuration Networks with Group Lasso Regularization
  • Greedy Stochastic Configuration Networks for Ill-Posed Problems
• Robust Data Analysis:
  • Intuitionistic Fuzzy Stochastic Configuration Networks
  • Weighted Deep Stochastic Configuration Networks Based on M-Estimator Functions
  • Noise Robust Regularized Deep Stochastic Configuration Networks
  • Robust Semi-Supervised Stochastic Configuration Network
• Deep Fusion Learning:
  • Deep Stochastic Configuration Networks Ensemble via Hyper-Parameter Optimization
  • Deep Stochastic Configuration Networks Ensemble via Boosting Negative Correlation Learning
  • Ensemble Intuitionistic Fuzzy Deep Stochastic Configuration Network
  • Stacked Deep Stochastic Configuration Networks with Multi-Level Feature Fusion
  • Stochastic Configuration Network with Long Short-Term Memory Feature Embedding
  • Book Review and Future Work

Readership: Introductory level graduate courses for numerical PDEs, finite element methods, finite difference methods, etc; Professors who are familiar with traditional numerical PDEs but also wish to cover modern data science applications would be very interested in such a book; Advanced undergraduate level courses for introducing numerical PDEs as well as data science.