README.md

CL_for_ImageSegmentation

repo for applying Continual Learning on Image Segmentation task.

Mainly consist of two parts:

• experience
  • experience on NAIVE, EWC, LFL, SI and GEM
  • report available on wandb link
• software engineering
  • a backend system for CL on Image Segmentation using UNet as basic model
  • Fastapi for restful api
  • Celery & Redis for Distributed Task Queue
  • Minio & Postgresql for dataset&checkpoint storage

Introduction

Three scenarios of Continual Learning

• Task-incremental learning: Sequentially learn to solve a number of distinct tasks
• Domain-incremental learning: Learn to solve the same problem in different contexts
• Class-incremental learning: Discriminate between incrementally observed classes

Our specific task is obviously not beyond the field of Domain-incremental learning

CL Learning Strategies Evaluation

Evaluation is based on U-Net and Avalanche.

please refer to experience for notebook and more details

report available on wandb link

Batch Domain Continual Learning

• Less-Forgetful Learning (LFL): paper | pdf

  • Less-forgetting Learning in Deep Neural Networks
  • Jung H, Ju J, Jung M, et al. Less-forgetting learning in deep neural networks[J]. arXiv preprint arXiv:1607.00122, 2016.

• Synaptic Intelligence (SI): paper | pdf

  • Continual Learning Through Synaptic Intelligence
  • Zenke F, Poole B, Ganguli S. Continual learning through synaptic intelligence[C]//International conference on machine learning. PMLR, 2017: 3987-3995.

• Elastic Weight Consolidation (EWC): paper | pdf

  • Overcoming catastrophic forgetting in neural networks
  • Kirkpatrick J, Pascanu R, Rabinowitz N, et al. Overcoming catastrophic forgetting in neural networks[J]. Proceedings of the national academy of sciences, 2017, 114(13): 3521-3526.

Online Domain Continual Learning

• Gradient Episodic Memory (GEM): paper | pdf
  • Gradient Episodic Memory for Continual Learning
  • Lopez-Paz D, Ranzato M A. Gradient episodic memory for continual learning[J]. Advances in neural information processing systems, 2017, 30.

Others

• Replay Buffer
• Selection strategies
• Loss Functions
• ...

SW engineering

please refer to sw_service for code and more details

ENV & Setup

please refer to setup for more details

Contact me

email: [email protected]