Explore ML detection methods for Saturn’s log detection module

[misilva73]

eta: 2023Q1

Description:

In order to deliver Saturn’s Milestone M0.2: Public L1 nodes, we had to go with a simple log detection system to catch fraudulent logs and operators. The reason for this was to lack real user data and the likelihood of having big shifts in data distribution after launch.

After the launch of the public network of L1s, we aim to collect data on those nodes and use that data to iterate and improve the log detection system.

Achieving this milestone involves:

• Coordinating with Saturn’s team to extract and label relevant log data
• Testing different ML strategies on the extracted data - anomaly detection, supervised learning, weak supervision, etc.
• If the experiments indicate that the tested strategies are better than the current model:
  • Making a proposal for changes;
  • Supporting the team with the model implementation.

The main deliverables are:

1. A report with the main results and final proposal for the model
2. Accompanying code for the reproducibility of all experiments
3. A packaged log detection model (in case of it being better than the current system)