Guanghan Wang
I am a fourth-year Engineering Science student in the Machine Intelligence major at the University of Toronto
Check my [Curriculum Vitae]
Welcome to contact me via:
I am a fourth-year Engineering Science student in the Machine Intelligence major at the University of Toronto
Check my [Curriculum Vitae]
Welcome to contact me via:
GPA: 3.93
The objective of the research is to determine if log files can predict the paths executed in an application. To begin, I collected logs and code coverage data using a fuzzer based on AFL. While collecting data, I constructed an LSTM neural network to predict the code region coverage of a log segment. By the end of the summer, the model had achieved an accuracy of approximately 90% on the openssh/wolfssh pair. Later on, I developed a decision tree to address the same problem and improved the accuracy to 99.7%.
Under the supervision of Prof. Nicolas Papaernot in the ClearHans Lab, I worked on audio adversarial machine learning and implemented a genetic algorithm to tackle the black box setting of speaker verification. During this process, I self-learned NumPy and TensorFlow. In the end, I achieved the goal of reducing the model's accuracy to below 1% through imperceptible genetic mutations of the input signal.
The result is part of the paper On the Exploitability of Audio Machine Learning Pipelines to Surreptitious Adversarial Examples