Importance Driven Continual Learning for Segmentation Across Domains

  title={Importance Driven Continual Learning for Segmentation Across Domains},
  author={Sinan {\"O}zg{\"u}r {\"O}zg{\"u}n and Anne-Marie Rickmann and Abhijit Guha Roy and Christian Wachinger},
The ability of neural networks to continuously learn and adapt to new tasks while retaining prior knowledge is crucial for many applications. However, current neural networks tend to forget previously learned tasks when trained on new ones, i.e., they suffer from Catastrophic Forgetting (CF). The objective of Continual Learning (CL) is to alleviate this problem, which is particularly relevant for medical applications, where it may not be feasible to store and access previously used sensitive… 
Adversarial Continual Learning for Multi-Domain Hippocampal Segmentation
This work proposes an architecture that leverages the simultaneous availability of two or more datasets to learn a disentanglement between the content and domain in an adversarial fashion, and takes inspiration from domain adaptation and combines it with continual learning for hippocampal segmentation in brain MRI.
Continual Hippocampus Segmentation with Transformers
An evaluation on hippocampus segmentation shows that Transformer mechanisms mitigate catastrophic forgetting for medical image segmentation compared to purely convolutional architectures, and demonstrates that regularising ViT modules should be done with caution.
What is Wrong with Continual Learning in Medical Image Segmentation?
This work proposes an evaluation framework that addresses both concerns, and introduces a fair multi-model benchmark that outperforms two popular continual learning methods for the task of T2-weighted MR prostate segmentation.
Continual Class Incremental Learning for CT Thoracic Segmentation
An adversarial continual learning segmentation approach (ACLSeg), which disentangles feature space into task-specific and task-invariant features enables preservation of performance on past tasks and effective acquisition of new knowledge is proposed.
Continual Active Learning Using Pseudo-Domains for Limited Labelling Resources and Changing Acquisition Characteristics
Results show that the proposed approach outperforms other active learning methods on a continuous data stream with domain shifts, and utilizes a rehearsal memory to avoid catastrophic forgetting while learning on new domains.
Recent Advances of Continual Learning in Computer Vision: An Overview
This paper presents a comprehensive review of the recent progress of continual learning in computer vision, grouped by their representative techniques, including regularization, knowledge distillation, memory, generative replay, parameter isolation, and a combination of the above techniques.
Domain Adaptation and Representation Transfer, and Distributed and Collaborative Learning: Second MICCAI Workshop, DART 2020, and First MICCAI Workshop, DCL 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 4–8, 2020, Proceedings
An auxiliary indicator function layer is designed to compress the network architecture via removing a decoding block, in which all individual responses are less than a given threshold α, that allows us to automatically identify and discard redundant decoding blocks without the loss of precision.


Towards continual learning in medical imaging
This work investigates continual learning of two segmentation tasks in brain MRI with neural networks and investigates elastic weight consolidation, a recently proposed method based on Fisher information that reduces catastrophic forgetting of the first task when a new one is learned.
Uncertainty-guided Continual Learning with Bayesian Neural Networks
Uncertainty-guided Continual Bayesian Neural Networks (UCB) is proposed, where the learning rate adapts according to the uncertainty defined in the probability distribution of the weights in networks.
Overcoming catastrophic forgetting in neural networks
It is shown that it is possible to overcome the limitation of connectionist models and train networks that can maintain expertise on tasks that they have not experienced for a long time and selectively slowing down learning on the weights important for previous tasks.
Continual learning: A comparative study on how to defy forgetting in classification tasks
This work focuses on task-incremental classification, where tasks arrive in a batch-like fashion, and are delineated by clear boundaries, and studies the influence of model capacity, weight decay and dropout regularization, and the order in which the tasks are presented, to compare methods in terms of required memory, computation time and storage.
Gradient Episodic Memory for Continual Learning
A model for continual learning, called Gradient Episodic Memory (GEM) is proposed that alleviates forgetting, while allowing beneficial transfer of knowledge to previous tasks.
Measuring Catastrophic Forgetting in Neural Networks
New metrics and benchmarks for directly comparing five different mechanisms designed to mitigate catastrophic forgetting in neural networks: regularization, ensembling, rehearsal, dual-memory, and sparse-coding are introduced.
Error Corrective Boosting for Learning Fully Convolutional Networks with Limited Data
SkipDeconv-Net (SD-Net), a new F-CNN architecture for brain segmentation that combines skip connections with the unpooling strategy for upsampling, is introduced and outperforms the latest state-of-the-art F- CNN models.
Reinforced Continual Learning
A novel approach for continual learning is proposed, which searches for the best neural architecture for each coming task via sophisticatedly designed reinforcement learning strategies, and which outperforms existing continual learning alternatives for deep networks.
PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruning
  • Arun Mallya, S. Lazebnik
  • Computer Science
    2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition
  • 2018
This paper is able to add three fine-grained classification tasks to a single ImageNet-trained VGG-16 network and achieve accuracies close to those of separately trained networks for each task.