This week, Team Banary focused on getting our project and deliverables ready for QRB 2. While we were previously struggling to reach convergence with the RNN model from William Dahl’s paper as well as our own model, we were able to fix the bugs and adjust the hyperparameters to make predictions on select data with significant accuracy. Along with breakthroughs in successfully executing the RNN models, we defined the prototype model and testing plan that will be shown in QRB 2. We reviewed our progress during this week’s liaison meeting. Although the liaisons were absent and the coach was dialing in while on the go, we received valuable insight on where to set our priorities. After the meeting, we assigned task items for our most important next steps, like data reformatting and system integration, picking up the pace as we head into March.

This week, the IPPD lecture was a team-wide work session. Banary spent three hours in NSC507 working together and exploring datasets and models. Our current approach is to fit an RNN on the executable’s assembly code, which can be decompiled from its binary. We are looking to reuse a publicly available model and build our product from it. After obtaining insights on the model’s architecture and data processing method, we will fit our own data into the model and measure its performance. While we have encountered some roadblocks, we will overcome it by seeking the help of our coach and liaisons. Meanwhile, we have kept eyes on alternative models, ready to switch paths if our current one does not work out.

This week, Banary presented our rendition of the “Design For X” approach. We chose to emphasize reliability, maintainability, and usability, which reflects the scope of work and how we defined our specifications and approach to the problem. We seek to make a product that reliably detects software vulnerabilities while following the best development practices to produce a maintainable and easy-to-use application. These ideas were well-received by our liaisons when we shared them during our meeting. Additionally, the team is steadily making progress on both paths noted in our project plan: making an RNN and adopting an existing model. We hope to produce sizable results to share during QRB 2.

This week Banary was reviewed by the Quality Review Board (QRB) on January 28, 2025 on the current status of CodeMine. The QRB consisted of Dr. Lattore, Dr. Abd-Elrahman, Dr. Patrick, and Dr. Resch (our coach). We had practiced presenting our slides the day before and the day of, feeling confident showcasing our goals and accomplishments. Banary was proud to show the kanban board we created outlining exactly what needs to be completed to ensure CodeMine completion. Following the presentation, we were given recommendations to improve our plan moving forward: include an explicit timeline, identify parallelism in tasks, search for additional models, and list our future risks in order of severity.

Finally we finished off the week with our weekly meeting with RTX! They were impressed with our organization and professionalism of our work tree and kanban board presented in QRB 1. Randall encouraged us to explore the paper and HuggingFace, as we will be in the coming week up to the February 14th deadline we have created.

Team Banary is excited to improve our work structure, show parallelism, and implement the model found in the research paper and those found in Huggingface to improve CodeMine.

Until next time. — Team Banary