ResNet10 Run Report

Summarized Findings

We introduced:

• label smoothing
• cosine annealing with warm restarts
• TrivialAugment

We had a misconfiguration in our cosine annealing config, meaning the rate was bouncing rapidly. Despite this, we yielded a significant improvement

Possible Improvements.

In our next implementation:

• We’ll introduce MixUp/CutMix
• EMA moving weight averages

Notebook

Please see the companion notebook. It contains:

• Training Confusion Matrix
• Validation Confusion Matrix
• Training Classification Report
• Validation Classification Report
• Training GradCam
• Validation GradCam