- This event has passed.
Muhammad Abu Nasar Rony Chowdhury
August 19 @ 2:00 pm - 3:00 pm MDT
Title: Improved Study of Side-Channel Attacks using Recurrent Neural Networks (RNNs)
Program: Master of Science in Computer Science
Advisor: Dr. John Stubban, Computer Science
Committee Members: Dr. Casey Kennington, Computer Science, and Dr. Marion Scheepers, Mathematics
With the rapid rise of technology, security concerns also increase to protect individual’s private and sensitive information. Almost every person is using a computing device and sharing their personal information to the third party or to the cloud. However no one knows, how his/ her information is taking care of. Hackers and intruders are always trying to break any security protocols and steal or leak confidential information to harm people around the world. To break cryptographic implementations on a computing system, side-channel attacks, are a prominent type of attacks. Side-channel attacks can be done in many forms: analyzing the power consumed by a device, counting time for computation, analyzing noises etc. Power analysis attacks and timing side-channel attacks are conducted by many adversaries by any means. These attacks are mainly based on information leaked by the hardware of a computing system or systems which make use of cryptographic protocols such as ATM’s and RFID smart-cards. If an attacker has physical access to the targeted device, he/she can recover sensitive information which is otherwise supposed to be hidden, for example, the devices key used for encryption. Like this, an adversary can recover all of the secret and sensitive information for destructive usage. These types of incidents can be lead people to indescribable loss. So, we need to increase the robustness of our current cryptographic implementations through introducing countermeasures in hardware and software.
Our main research objective is to conduct a thorough, comprehensive and in-depth study of the application of recurrent neural networks (RNNs) in the context of side-channel attacks.
The difference between our work and previous work in this field is that for the first time, we will use Recurrent Neural Networks (RNNs) in differential power analysis: side-channel attacks. Earlier work only showed convolutional neural networks performance in attacking cryptographic implementations, whereas, we will show RNNs’ performance and we will also compare between the performance of RNNs and CNNs. Through RNN network, we will try to detect underlying cryptographic algorithm of the device. And we will predict the key using program which is developed for cryptanalysis of cryptographic algorithms.
To achieve our objective, we are trying to give answers to the following research questions:
1. Is it possible to detect cryptographic algorithms (DES, AES etc.) through our proposed RNN model?
2. What should be the size of the datasets to decide our RNN model a good one and how many hidden layers are required to train all the datasets efficiently?
3. Is it possible to achieve greater efficiency by using RNNs instead of using Convolutional Neural Networks (CNNs) or Machine Learning Techniques in side-channel attacks?
In order to answer our research questions, we use two cryptographic algorithms: DES and AES for conducting our research. We apply different modeling algorithm including the Long Short-Term Memory (LSTM) technique to model our neural network.