Computer Vision News - April 2023

45 Camila González Moving beyond catastrophic forgetting to adapt systems over time Of course, detecting failures is only the first step. We seek to train models that are reliable for a large portion of the data. Domain generalization and adaptation only increase robustness to a certain extent. As time goes on, models can only maintain acceptable performance if they learn from newly acquired cases. The goal of continual learning is to adapt to changes in the environment without forgetting previous knowledge. One practical strategy to approach this relies on expansion , wherebymultiple parametrizations of themodel are maintainedand themost appropriateone is selected during inference. We present two expansion-based methods that do not rely on information regarding when the data distribution changes. Overcoming practical hurdles Even when appropriate mechanisms are in place to fail safely and accumulate knowledge over time, this will only translate to clinical usage insofar as the regulatory framework allows it. Current regulations in the USA and European Union only authorize locked systems that do not learn post-deployment. Fortunately, regulatory bodies are noting the need for a modern lifecycle regulatory approach . We review these efforts, along with other practical aspects of developing systems that learn through their lifecycle. We are finally at a stage where healthcare professionals and regulators embrace deep learning. The number of commercially available diagnostic radiology systems is skyrocketing. This opens up our chance - and responsibility - to show that these systems can be safe and effective throughout their lifespan. RIGIDITY PLASTICITY RESOURCE USE Regularization Pseudo-reh. Expansion Sparsity CONTINUAL LEARNING STRATEGIES CRITERIA Memory Performance Task boundaries RIGID SMOOTH Task labels YES NO TASK LABELS Not known Known