Authors:
Ajay Shrestha
and
Ausif Mahmood
Affiliation:
Department of Computer Science and Engineering, University of Bridgeport, 126 Park Ave, Bridgeport, CT 06604 and U.S.A.
Keyword(s):
Siamese Networks, Importance Sampling, Dataset Optimization, Convolution Neural Networks.
Related
Ontology
Subjects/Areas/Topics:
AI and Creativity
;
Artificial Intelligence
;
Biomedical Engineering
;
Biomedical Signal Processing
;
Computational Intelligence
;
Data Manipulation
;
Evolutionary Computing
;
Health Engineering and Technology Applications
;
Human-Computer Interaction
;
Knowledge Discovery and Information Retrieval
;
Knowledge-Based Systems
;
Machine Learning
;
Methodologies and Methods
;
Neural Networks
;
Neurocomputing
;
Neurotechnology, Electronics and Informatics
;
Pattern Recognition
;
Physiological Computing Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
;
Symbolic Systems
;
Theory and Methods
Abstract:
The accuracy of machine learning (ML) model is determined to a great extent by its training dataset. Yet the dataset optimization is often not the center of the focus to improve ML models. Datasets used in the training process can have a huge impact on the convergence of the training process and accuracy of the models. In this paper, we propose and implement importance sampling, a Monte Carlo method for variance reduction on training siamese networks to improve the accuracy of the image recognition. We demonstrate empirically that our approach can achieve improvement in training and testing errors on MNIST dataset compared to training when importance sampling is not used. Unlike standard convolution neural networks (CNN), siamese networks scale efficiently when the number of classes for image recognition increases. This paper is the first known attempt to combine importance sampling with siamese network and shows its effectiveness towards getting better accuracy.