What project milestones did you accomplish this week?
Conference Paper
This week I finished writing my part of a conference paper that is being reviewed by Dr. Thulasiraman. It contains a summary of all my work over the last few months involving feature extraction from 5G and Modbus PCAP files and the training and tuning of basic autoencoder models. This conference paper provides the basis for my current work.
Model Configurations
This week I created a convolutional autoencoder model. This was the final model I needed to build. I already tested it with the original feature representation (ICMP flood) using scaled and unscaled data.
Feature Representations
I finished aggregating results for my “novel flows” idea using the basic autoencoder model. Rather than using “synthetic flows” to test my models (as before), I excluded a couple of real flows from training and marked their subflows as anomalous. The partitioning of the flows is also different from the previous work (forward flows vs. inward flows).
What is your plan for next week?
Over the next couple days, I am going to finish testing my models with all the feature representations I have. Three models are being compared for anomaly detection: Basic autoencoder, convolutional autoencoder, and KMeans. Feature representations include flows (forward and inward), MITM, and query responses. Results will be aggregated and sent to Dr. Thulasiraman for review. This will conclude my initial testing, although I do not expect any breakthroughs.
On Friday, Raytheon network engineers will be meeting with our team to discuss steps forward. We will be discussing the network architecture and concept of operations (CONOPS). Hopefully this results in new ideas for feature representations. I am going to ask questions about query response values I found in the Modbus data. I am also going to try to get an update on when we can expect new feature sets.
No comments:
Post a Comment