Contrastive Unlearning with MOCO

This repository implements a contrastive unlearning approach using MOCO (Momentum Contrast) framework. The main implementation is in unlearn.py, which contains:

• MOCO_Unlearn class that handles the contrastive unlearning process:
  • Maintains a queue of feature embeddings and corresponding labels
  • Implements contrastive loss between unlearning samples and queue samples
  • Combines contrastive loss with cross-entropy loss for retained samples

The core library is shared with the contrastive unlearning implementation.