A/2, Jahurul Islam Avenue
Jahurul Islam City, Aftabnagar
**Currently on leave to pursue a PhD degree in Computing at Macquarie University, Sydney, Australia.
For current researches and past supervisions, please see the RESEARCH INTEREST section.
Dept. of Computer Science and Engineering, East West University, Dhaka.
Bangladesh Army International University of Science and Technology (BAIUST), Comilla
International University of Business Agriculture and Technology (IUBAT), Dhaka
Research interests include - Machine/ Deep Learning, Data Science, Data Security, Software Engineering, Web Application.
Detecting Adversarial Purterbation to Defend Adversarial Attacks on Text Data: This research aims at exploring adversarial attacks by generating adversarial texts that deceive Natural Language Processing (NLP) models into producing the wrong prediction and propose defence against such attack in order to improve the model's robustness and preserving data privacy. Our primary focus is on the NLP models that predict an author’s individual information such as age and gender. Our machine learning algorithms on text dataset in both centralized and decentralized/federated learning settings to ensure data privacy and resilience against adversarial attacks.
Adversarial Training-resistant Machine Learning models (on-going): This research aims to use adversarial training in MLAAS models implemented both in centralized and decentralized/federated learning settings to ensure data privacy and resilience against adversarial attacks. The objectives of this research are: i) Provide a comprehensive analysis of different adversarial attacks on supervised, unsupervised3and re-inforced ML algorithms in both centralized and federated MLAAS models. ii) Investigate the robustness of MLAAS models against black-box adversarial attacks by using the adversarial training method and including adversarial examples in the training data before training the actual ML algorithm. iii) Use the adversarial training method to detect and remove malicious updates to the model in federated settings against white-box attacks. iv) Extend the robust adversarially trained MLAAS model to enable secure aggregation of data collected from multiple sources/data owners by incorporating adversarial examples into them while ensuring optimal output from the ML model trained on the aggregated data. v) Study the optimization problem to have an optimal balance between privacy (resilience against adversarial attacks) and ML performance.
Information Leakage in Machine Learning Models: Machine Learning (ML) techniques are used by most data-driven organisations to extract insights. Machine-learning-as-a-service (MLaaS), where models are trained on potentially sensitive user data and then queried by external parties are becoming a reality. However, recently, these systems have been shown to be vulnerable to Membership Inference Attacks (MIA), where a target’s data can be inferred to belong or not to the training data. While the key factors for the success ofMIA have not been fully understood, existing defence mechanisms only consider the model-specific properties. We investigate the impact of both the data and ML model properties on the vulnerability of ML techniques to MIA. Paper link: https://arxiv.org/pdf/2002.06856.pdf
An Improved String Matching Algorithm for Efficient Text Mining: String searching or matching algorithms play an important role in many cases where a certain text or string is required to match with other texts. With the increase of data storing and usage on a large scale, it is very important to find a faster alternative. A new and efficient algorithm will be developed analyzing existing pattern matching algorithms (e.g. Knuth-Morris-Pratt (KMP) Pattern matching, Boyer-Moore string search algorithm etc.). Supervisee: M. O. Al-Faruk, K. M. Hussain, M. A. Shahriar. Published journal paper link: https://doi.org/10.1007/s41870-019-00371-1
Ecommerce Web Data Analytics and Prediction: Web application using MVC and Entity Framework. This is an e-commerce project that would make suggestions on how much the future inventory should update based on analyzing past customer clicks and buying habits. Also, when a customer searches for products, this app would let the manager set the percentage of noise from the lowest viewed products to be injected among the highest viewed products. This would ensure a distributed amount of product appearances in the product list. Supervisee: Jubair Hridoy, East West University Project Url: https://github.com/hridoy29/EcommerceProject
Spark- IOT Based Smart Sparking System: Using Arduino Mega, PHP & MySQL database. This system simulates a web-based system from where the user will be able to get the latest information on the location of available parking spots using LDR sensors. Also, there is a password protected gate locking system for the security of the parked vehicles. Supervisee: Shibly Sirajee & Mashuk Kawser, East West University. Project Url: https://github.com/shiblysirajee/Automatic-Car-Parking-Search
TRAVELOUS- An Information Portal for Tourists: Web application using MVC & Entity Framework. This portal is a common platform for tourists and vendors to share knowledge on different tourist spots, attractions, cultures, and vendor services in the area. Supervisee: Samia Bari, East West University. Project Url: https://github.com/samiabari/Travelous.git
S. M. Tonni, S. Parvin, A. Gawanmeh and J. Jackson, “Protecting Big Data through Microaggregation Technique for Secured Cyber-Physical Systems,” in “Cyber-Physical Systems for Next Generation Networks”, IGI Global, Hershey, PA, USA, May 2018, pp. 99-120.
S. M. Tonni, M. Z. Rahman, S. Parvin and A. Gawanmeh, “Securing Big Data Efficiently through Microaggregation Technique,” 2017 IEEE 37th International Conference on Distributed Computing Systems Workshops (ICDCSW), Atlanta, GA, USA, 2017, pp. 125-130. doi: 10.1109/ICDCSW.2017.65,
M. O. Al-Faruk K. M. Hussain, M. A. Shahriar and S. M. Tonni, “BFM: a forward-backwards string matching algorithm with improved shifting for information retrieval”, In International Journal of Information Technology, Oct. 2019, issn: 2511-2112, doi:10.1007/s41870-019-00371-1.
Tonni, S. M., Vatsalan, D., Farokhi, F., Kaafar, D., Lu, Z., & Tangari, G. "Data and model dependencies of membership inference attack." arXiv preprint arXiv:2002.06856 (2020).