Towards Robust Feature Learning with t-vFM Similarity for Continual Learning
This work addresses data imbalance challenges in continual learning for image classification, though it appears incremental as it modifies an existing loss function.
The paper tackled the problem of learning robust representations in continual learning by proposing a new similarity metric, t-vFM, to replace cosine similarity in supervised contrastive loss, resulting in improved performance on the Seq-CIFAR-10 dataset that outperforms recent baselines.
Continual learning has been developed using standard supervised contrastive loss from the perspective of feature learning. Due to the data imbalance during the training, there are still challenges in learning better representations. In this work, we suggest using a different similarity metric instead of cosine similarity in supervised contrastive loss in order to learn more robust representations. We validate the our method on one of the image classification datasets Seq-CIFAR-10 and the results outperform recent continual learning baselines.