Publications

A Deep Incremental Learning Framework for Predicting Covid-19 by using Incoming Stream X-ray Images of Chest

Published in 7th International Conference on Technology Development in Iranian Electrical Engineering, 2022

The COVID-19 epidemic has erupted in more than 150 nations around the world. One of the quickest ways to diagnose patients is to use radiography and radiology images to detect this disease. As the disease has not yet been eradicated, the number of these images is increasing daily and the dataset is constantly growing. In our framework, Covid -Stream, Images are entered into the framework as a stream of data. The proposed framework consists of two main parts. in, transfer learning phase features are extracted from these batch images using Keras library. Then incremental learning is applied to predict and evaluate COVID and non-COVID images using Creme library. Incremental learning plays an important role in this framework because it is not possible to process and fit all data into the memory. The proposed framework is tasted on a public CXR dataset (named COVID X-ray-5k) containing different chest abnormalities, and the proposed method achieved an accuracy of 0.86%. It also achieved a highly competitive performance while significantly reducing the training and computational burden. The proposed framework can solve real-world big datasets scalability issues.

Recommended citation: Sadeghi-Nasab, Alireza and Shakoor, Mohammad Hossein,1401,A Deep Incremental Learning Framework for Predicting Covid-19 by using Incoming Stream X-ray Images of Chest,7th International Conference on Technology Development in Iranian Electrical Engineering,Tehran,https://civilica.com/doc/1492958 https://civilica.com/doc/1492958

A New Fast Framework for Anonymizing IoT Stream Data

Published in 2021 5th International Conference on Internet of Things and Applications (IoT), 2021

The Internet of Things (IoT) plays an important role in human life today. Millions of devices generate and transmit vast amounts of data. Exploring this data without compromising privacy practices may expose to risk of users’ identities. One of the measures used to protect data privacy is anonymity methods. IoT data anonymization is not possible using traditional methods because such data, unlike database data, are not static and are very large. In this paper, we propose a new framework that can anonymize the received stream data by considering their expiration time. This anonymization is performed using a new clustering method using a streaming data processing engine. The introduced clustering method has a significant effect on reducing data delay. It supports both numerical and categorical data types too. Also, merging remaining clusters at the end of the method has minimized information loss. Comparing the performance results of the introduced method with similar methods shows that the proposed method performs better in terms of information loss and data delay.

Recommended citation: Nasab ARS, Ghaffarian H. A New Fast Framework for Anonymizing IoT Stream Data. In2021 5th International Conference on Internet of Things and Applications (IoT) 2021 May 19 (pp. 1-5). IEEE. https://ieeexplore.ieee.org/document/9469718