I’ve finished lesson 10 and I am slightly confused. We’ve started with a simple model (Linear, ReLu, Linear) and high accuracy in just one epoch (0.9+) and ended up with a convolutional network (more complex), batch normalisation, learning rate schedulers and ended up at accuracy of 0.97.
The accuracy I got with the first model after 2 epochs was 0.97, which is very close to 1. Is there are better dataset (that I can train on from scratch) to show the power difference between a linear layer and convolutions or the job is to chase these 1 or 2% better accuracy?