Broad learning vs. deep learning

• Main Author: Gu, Haowei
• Other Authors: Wang Lipo
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/152974
• Institution: Nanyang Technological University

In recent years, deep learning has been used in many kinds of industrial fields and saw dramatic breakthroughs. Many projects have been applied in real life, such as face recognition, natural language processing. Training deep learning models requires high levels of computer hardware and training processes will cost a lot of time. Once the dataset is changed, the current network cannot get a good recognition effect, data scientists have to change the network structure and train it again. Broad learning is a good method to alternate deep learning because broad learning only changes some parameters in the current broad learning model and so some simple calculations when the dataset is changed. Compared with deep learning, broad learning costs less time and does not need high computer hardware, which is considered as a suitable model to replace the Deep Learning model. In this project, the result of comparisons between Broad Learning model and Deep Learning model with Fashion MNIST and noisy MNIST dataset which illustrates the effect of Broad Learning model and whether it would be an alternative method for Deep Learning. The result shows that although the BLS system does not get the highest accuracy, the costs time on training processing is the least, and BLS system gets the best result when the dataset is noisy.