Inspiration: My computer security research at RIT. I need a tool to detect anomalous traffic in Networks-on-a-Chip (NoC).
What it does: It's a Conditional GAN (cGAN). It learns "normal" NoC traffic from simulation CSVs so it can spot "abnormal" (malicious) traffic.
How we built it: The biggest "hack" was building a data pipeline in Python/Pandas to parse complex simulation CSVs. This pipeline feeds a cGAN built in Keras/TensorFlow.
Challenges we ran into: Figuring out how to parse the exact 16x16 matrix and specific benchmark/core labels from different lines in the same CSV file.
Accomplishments that we're proud of: Getting the data pipeline to successfully read, shape, and normalize data from many CSVs into a single, clean dataset.
What we learned: How to build an ML pipeline, modify a Keras tutorial for a real dataset, and normalize data for a tanh activation.
What's next for NoC Sentry GAN: This is a 2D "snapshot" detector. The next step is a 3D spatio-temporal GAN to analyze sequences of traffic (like a video) for more complex anomalies.
Log in or sign up for Devpost to join the conversation.