Research Project Title:
Generating Adversarial Computer Programs Using Optimized Obfuscations
abstract:Machine learning models for code related tasks are increasingly being used in the software engineering field. These models can often be a target for adversarial attacks which aim to manipulate the model predictions by providing deceptive inputs. As a result, it is important to have a way to measure how robust these models are to this type of attacks. In this project we will work on an optimization approach to generate adversarial computer programs which we will use to evaluate the robustness of machine learning models used for the code summarization task. Moreover, the generated computer programs can be used for adversarial training of the same models in order to increase their robustness to adversarial attacks.
I am participating in this SuperUROP because I am interested in machine learning and I want to gain more research experience in this field. I hope to get more knowledge related to my major and see what working on real research projects looks like.