Please use this identifier to cite or link to this item:
https://hdl.handle.net/10356/153483
Title: | DNN-FES & Fole : objective loss estimator for integrating deep neural-network into fuzzy systems | Authors: | Dandapath, Soham | Keywords: | Engineering::Computer science and engineering | Issue Date: | 2021 | Publisher: | Nanyang Technological University | Source: | Dandapath, S. (2021). DNN-FES & Fole : objective loss estimator for integrating deep neural-network into fuzzy systems. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/153483 | Project: | SCSE20-0786 | Abstract: | For many years, deep learning has been a popular study area. Its effectiveness in a variety of applications has led to immense research in this sector, leading to several neural network architecture that are as accurate as humans. Despite possessing human-like accuracy, deep neural networks mostly lack human like cognition which explains the reasoning behind their decisions, making them akin to a black box. Fuzzy Systems, on the other hand, are based on human-like reasoning, inspired by the brain's hippocampus. However, they are vulnerable to outliers and do not generalize well unlike the deep neural networks making them quite fragile. Attempts to merge these two fields are nothing new. However, owing to the use of non-differentiable membership functions in fuzzy systems, neuro fuzzy systems based on the Mamdani inference scheme have only been able to adapt the structure of the neural network and not the learning mechanism. Deep Neural Network - Fuzzy Embedded System (DNN-FES) is a unique architecture based on the Mamdani inference scheme. It aims to replace a normal T-norm with a neural network that learns the inference process using learning mechanism of the neural network. We also introduce Fuzzy Object Loss Estimator (FOLE) to leverage backpropagation algorithm of neural networks into the neuro fuzzy system. FOLE is an objective loss function in the sense that it not only tries to minimize the error incurred from the target and predicted value, but also the loss incurred in making inadmissible or unjustifiable rule. In most of the benchmarking tests, DNN-FES has proved to produce state-of-the-art results. For the application of DNN-FES in stock market prediction, the results have shown quite promising results making it a potential candidate for its usage in stock market trading and portfolio balancing. | URI: | https://hdl.handle.net/10356/153483 | Schools: | School of Computer Science and Engineering | Fulltext Permission: | restricted | Fulltext Availability: | With Fulltext |
Appears in Collections: | SCSE Student Reports (FYP/IA/PA/PI) |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Soham_Dandapath Final Report.pdf Restricted Access | Undergraduate Thesis - Soham Dandapath | 5.31 MB | Adobe PDF | View/Open |
Page view(s)
310
Updated on Mar 28, 2024
Download(s) 50
32
Updated on Mar 28, 2024
Google ScholarTM
Check
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.