Visualising and Explaining CL Model while learning new classes and tasks

Hello everyone,
We are currently working on Continual learning methods in the Network Intrusion domain. Typical Network Intrusion datasets have a heavy class imbalance. (1 Normal class constitutes nearly 75-80% of the total data and the remaining all classes constitute nearly 20-25% of data).
Our Training Setting lin CL framework will typically be divided into a sequence of tasks where each task will contain some classes of either normal/attack. We want to visually see how the model learns and changes after learning each class and how the model changes after learning all classes in a particular task and we also wanted to visualize how the model distinguishes b/w tasks. Any resources involving these tools would be very much helpful.

And also I didn’t find any CL algorithms very specific to these kinds of datasets to compare the results and the very specific metrics for this kind of dataset. Any papers would be helpful!!

1 Like

Hi @akashtadwai! You can take a look at Avalanche. We have a couple of metrics you can use there and it is rather easy to add new visualization method on Tensorboard or Wandb!