Applications of Continual Learning Settings in Medical Imaging

Author: Dhananjay Nahata

Virtual Room:

Date & time: 22/09/2021 – 11:00 h

Session name: Continual Leraning

Supervisor: Joost Van De Weijer


The application of Artificial Intelligence (AI) has tremendously impacted our lives, which has led to rapid demand and usage of technologies in various domains such as classification, object detection (OD) and recognition, etc. Deep neural networks (DNNs) had already achieved a significant milestone in tackling the classification problems in classifying a certain task, where it generally learns from well-defined phase, thus acquiring knowledge of that task only on which it is trained, but these architectures face the issue of the catastrophic forgetting, where it fails to extend the knowledge of the previous task when trained on a new task. Incremental learning (IL) tries to address this issue by accommodating the knowledge of the previous tasks and current tasks continually, without training the model from the scratch for different tasks. In this work, we are trying to address classification problem task incremental learning on the medical images of breast cancer and rectal colon using various state-of-the-art methods, where we define certain tasks and the network tries to learn each task sequentially.


– President: Josep R. Casas(UPC)
– Secretary: Jorge Bernal(UAB)
– Vocal: Montse Pardàs(UPC)