1. DeepMem: ML Models as storage channels and their (mis-)applications: Machine learning (ML) models are overparameterized to support generality and avoid overfitting. In this paper, we propose a novel information theoretic perspective of the problem; we consider the ML model as a storage channel with a capacity that increases with overparameterization. Specifically, we consider a sender that embeds arbitrary information in the model at training time, which can be extracted by a receiver with a black-box access to the deployed model. We proposed black-box write and read primitives that allow the attacker to: (i) store data in an optimized way within the model by augmenting the training data at the transmitter side, and (ii) to read it by querying the model after it is deployed. We also analyze the detectability of the writing primitive and consider a new version of the problem which takes information storage covertness into account. Specifically, to obtain storage covertness, we introduce a new constraint such that the data augmentation used for the write primitives minimizes the distribution shift with the initial (baseline task) distribution. Moreover, we also develop optimizations to improve the capacity, including a novel ML-specific substitution based error correction protocol. We believe that the proposed modeling of the problem offers new tools to better understand and mitigate potential vulnerabilities of ML, especially in the context of increasingly large models. This work was done under the expert supervision of my Ph.D. supervisor Prof. Nael Abu-Ghazaleh from UCR and Co-advised by Prof. Ihsen Alouani from QUEEN'S UNIVERSITY BELFAST.
2. Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attacks: This paper surveys research in the emerging interdisciplinary field of adversarial attacks on LLMs, a subfield of trustworthy ML, combining the perspectives of Natural Language Processing and Security. In this survey, we first provide an overview of large language models, describe their safety alignment, and categorize existing research based on various learning structures: textual-only attacks, multi-modal attacks, and additional attack methods specifically targeting complex systems, such as federated learning or multi-agent systems. We also offer comprehensive remarks on works that focus on the fundamental sources of vulnerabilities and potential defenses. To make this field more accessible to newcomers, we present a systematic review of existing works, a structured typology of adversarial attack concepts, and additional resources, including slides for presentations on related topics at the 62nd Annual Meeting of the Association for Computational Linguistics (ACL'24).
34. A Novel Approach to Blockchain-based Digital Identity System: This work was done for my undergraduate thesis under the expert supervision of Prof. Dr. Md. Shohrab Hossain and it has been accepted in FICC, 2020 and the proceedings of FICC, 2020 has been published in the series "Advances in Intelligent Systems and Computing (AISC, volume 1129)" of Springer. This work is based on Blockchain technology which is a secured platform to store confidential information. An individual creates his digital identity using his bio-information and a predefined private key. A citizen can perform all activities inside his country without any carriable paper documents or identity card. We implemented the proposed Digital identity creation Algorithm using Etherium smart contract, Online Remix IDE and Solidity,an object-oriented, high-level programming language for implementing smart contracts.
4. Towards Perceiving and Resolving the Impediments to Reporting For the Developing Countries: A platform for online reporting system in Bangladesh. This work was done during my undergraduate studies under the supervision of Prof. Dr. A.B.M Alim Al Islam Razi and was published in 2017 4th International Conference on Networking, Systems and Security (NSysS). A one-stop generic reporting platform can be a useful collaborative tool between common people and responsible authority in a developing country. UReporter is a platform for common people to report anonymously any unusual things or crimes happened at any place. Common people will also get the feedback from the appropriate authority using this platform. We have developed a website regarding Ureporter. Visit our facebook page and official site.
5. Image Steganography using Multiple LSB substitution and Pixel Randomization using Stern-Brocot Sequence: This work was done during my undergraduate studies under my Thesis Supervisor Prof. Dr. Md. Shohrab Hossain and this paper has been accepted in FICC, 2020 and was published in the series "Advances in Intelligent Systems and Computing (AISC, volume 1129)" of Springer. In essence A hidden message is embedded within an image in a secured way. The pixel within the image was selected randomly so that no intruder can extract the actual message without knowing the sequence of pixel selection and the encryption-decryption key. The proposed algorithm has been implemented using JAVA language.
6. A Novel Approach to Manage Ownership and VAT Using Blockchain-Based Digital Identity : In this research paper, a Blockchain based Ownership and Value Added Tax (VAT) management system has been proposed depending on a digital identity system that is implemented on blockchain technology using individuals' biometric information. The Ownership system along with VAT management was implemented using Ethereum smart contract. The experimental results of testing our system indicate that an intruder cannot perform any modification illegally to ownership data in the system. Any adversarial attempt is aborted instantly and thereby the security of citizen's ownership information is ensured. Moreover, appropriate amount of VAT is automatically assigned to the owner while changing the ownership of suitable products.
1. Techlance : In this project, I tried to develop a system where all freelancer can find the request of their desired customers to do a particular job and can accept their desired offer. Here, Freelancer will do his required job within client specified days and submit his completed work on this website. Then the customer will receive his work in this website and completes the payment online. JavaSwing, HTML, CSS, PHP, MySql was used for building this system. This work was done during my 4th year undergraduate studies for my Software Development sessional project. (Github Link)
2. Language Learning Quiz Game: In our software Engineering sessional, I made a question-answer platform named language learning system. This work was done during my 3rd year undergraduate studies. I used CodeIgniter as framework and Javascript, HTML, CSS for making this quiz game. One could play Quiz on a preferred language to answer those in another language to check the selected language proficiency. (Github Link)
3. Ticket reservation System: I used JavaFX for making this system where a passengers could book his ticket online. There was also option to buy the ticket and cancel the booked ticket. This work was done during my 2nd year undergraduate studies. (Github Link)
4. Snake Game: I made this game using Assembly language for our Assembly Language Sessional project. This project was done in my undergraduate Level-2 ,Term-1 period. (Github Link)
5. DX Ball Game: The Project was done in the period of my undergraduate first year. IGraphics was used for making the game in my Level-1 ,Term-1. (Github Link)
6. Cost Efficient Touch Sensor: This Project was done when I was a student of Level-3 Term-2 in BUET. It aims to develop a cost efficient touch sensor. For this project we have used 16:1 mux which has 4 selector which can enable one pin at the mux at a time. In order to implement touch sensor we have used open circuit for each dot of the dot matrix. Each open circuit is represented by two pins . After putting on a gloves with a small piece of metal on the verge of finger . when the two pins will be touched together using the gloves metal on it , then that particular circuit is full. As a result corresponding dot of the dot matrix will be bright. Thus each dot of the dot matrix will be bright if we touch each pair of pins. (Github Link)