PLCL: Partial-Label Continual Learning

PLCL (Partial-Label Continual Learning) is a research project focusing on addressing challenges in partial-label continual learning. This project introduces an innovative weakly supervised learning algorithm that achieves performance comparable to fully supervised methods, providing a benchmark for future research in this domain.

Overview

• Objective: Develop an effective weakly supervised algorithm to tackle key issues in partial-label continual learning.
• Key Contributions:
  • Proposed a benchmark algorithm for partial-label continual learning.
  • Achieved performance close to fully supervised methods in weakly supervised scenarios.
  • Provided a reproducible framework and comprehensive experimental results for continual learning research.

Project Structure

• Core Implementation:
  • PLCL.py: Main algorithm implementation.
  • model.py & resnet.py: Define the model architecture and backbone network.
  • experiment.py: Experimental script for training and evaluation.
• Utilities:
  • utils/: Tools for logging and auxiliary algorithms.
• Documentation:
  • report.pdf: Detailed project report.

Quick Start

1. Clone the Repository:

   git clone https://github.com/KevinCarpricorn/PLCL.git
   cd PLCL

2. Run the Experiment:

     python experiment.py

Modify parameters in experiment.py as needed to customize the experimental setup.

Results

• The proposed algorithm demonstrates robust performance in partial-label continual learning tasks.
• Experimental findings indicate comparable results to fully supervised methods in weakly supervised settings.

For detailed results and analysis, refer to report.pdf.

License

This project is licensed under the MIT License.