Grad Projects
1. Debopriya Roy Dipta, Seonghun Son, “Dynamic Frequency-Based Side-Channel Attack against Modern Sandbox Environments”, CPR E 681: Advanced Topics in Computer Architecture, Iowa State University, 2023.
In this project, we exploit Dynamic Voltage and Frequency Scaling (DVFS) based side-channel to leak information about containerized apps that are running on the host laptop. In the new era of serverless computing, cloud providers run functions inside isolated containers protected with extra layers of virtualization developed by the owner itself. For example, AWS launched Firecracker, which allows the user to create multiple microVM with a faster booting-up feature. Inside each microVM, the user can run containers securely. Although this microVM is completely protected against different micro-architectural attacks, we explore the vulnerability against the DVFS-based side-channel attack. According to our study, we found that a DVFS-based side-channel attack is capable of leaking information about the containerized apps running inside the microVM of the firecracker tool.
2. Debopriya Roy Dipta, “Comparison of Cache Replacement Policies and realize its impact on microarchitectural attacks using Gem5 Simulator”, CPR E 581: Computer Architecture, Iowa State University, 2021.
In this project, we carried out a detailed comparative analysis among some of the common and efficient cache replacement policies with an aim to study their individual effectiveness for different cache configurations, which is performed by using the gem5 simulator. The evaluation is carried out by using benchmarks with varying workloads to determine the performance metrics individually. In this current study, we evaluated the performance of Random Replacement (RR), Least Recently Used (LRU), and First in First Out (FIFO) replacement policies with benchmark Nqueens and BFS. In addition, this project also intends to find out the impacts of the above-mentioned replacement policies on time-driven cache side-channel attacks’ performance in terms of success rate.
3. Debopriya Roy Dipta, “Website Fingerprinting by leveraging Intel RAPL-induced side channel.”, EE 526: Deep learning theory and practice, Iowa State University, 2021
In this project, 1) we demonstrated the possibilities of using RAPL readings to fingerprint websites. Unlike the previous work on using power traces for website fingerprinting which requires external instrumentation and hardware modification, the RAPL-based technique is purely in software based. 2) We evaluated the RAPL-based website fingerprinting technique using Chrome browser on Linux machine. The results show that this CNN based fingerprinting method is highly effective in terms of the regular network, up to 92% testing accuracy can be achieved 3) We investigated the common pattern among the energy traces while browsing the same website in browser and utilized a convolutional neural network model to recognize the fingerprint of a visited website.
Undergraduate Thesis
Isolated Speech Recognition System Using Convolutional Neural Network
Thesis supervisor:
Dr. Md. Mahbub Hasan
Professor
Department of Electrical and Electronic Engineering
Khulna University of Engineering and Technology, Bangladesh.
Thesis Keywords: CNN, Speech Recognition, Feature extraction, MFCC, PLP.
A Convolutional Neural Network (CNN) has been proposed for isolated speech recognition. For embedding each audio input Mel Frequency Cepstral Coefficients (MFCC) and Perceptive Linear Prediction (PLP) Coefficients have been utilized. The model has been justified using two different dataset. One is for English and the other is for Bangla. For CNN, three convolution layer has been used. In this three layers ‘ReLu’ is used as the activation function for the neurons. Activation function ‘tanh’ is also used for making a comparison between this two. As a dataset, we have 20 isolated speech or words spoken by 1700 people for each words. So, for English we have total 34000 audio input. For Bangla we have five isolated word spoken by 100 people each. So for Bangla dataset we have total 500 audio input. After doing MFCC and PLP, we have fed the Feature vector to our CNN model separately. The highest accuracy that we have found for English dataset is around 85% and for Bangla dataset the accuracy is around 82%.
Software Dependencies:
In the total thesis the software that has been utilized so far are mentioned below:
-
Python 3.5.2 (Anaconda3 4.2.0 64 bit)
-
Tensor Flow with GPU support
-
API: Keras
-
Praat for Data collection
-
Matlab.
Data acquisition with RODE microphone using Scarlett 2i2 interface (For Bangla Speech)
Results
Undergraduate Project
FACE RECOGNITION BASED SECURITY SYSTEM Utilizing PCA Algorithm
Duration: 6 months
The main objective of this project is to ensure the security using a face recognition system. For face recognition, a model has been built which has been trained with a preformed database of allowed persons. The task of this recognition model is to recognize any incoming test image and verifying the identity of that person. The decision made by the face recognizing model whether the test image is in the preformed database or an intruder, is sent to the arduino for doing the hardware task. The hardware part contains a Liquid Crystal Display (LCD), for showing the outcome or the decision that has been made by the face recognizing model and also a servo motor for automatically controlling the lock of the door. For capturing a test image webcam has been used. Hence, the entire project can be divided into two parts. The first one is the software part, in where we have built an efficient model for recognizing face of any test person. The second part is the hardware part, which acts according to the results of our software part. For face recognition, we have used PCA based algorithm with appropriate feature extraction. The code for hardware part has been written down in Arduino IDE. Next, we have interfaced Arduino IDE with MATLAB so that they can communicate with each other by sending and receiving data.
CIRCUIT DIAGRAM FOR THE HARDWARE PART
Simulation Result of the Software Part
