Bayesian Learning-driven Prototypical Contrastive Loss for Class-Incremental Learning

TL;DR

BLCL addresses class-incremental learning by dynamically expanding task-specific blocks and balancing cross-entropy with a prototypical contrastive loss through Bayesian uncertainty weighting. The method builds on MEMO, introducing dynamic specialization and a Bayesian objective that learns $\sigma_1$ and $\sigma_2$ to adapt loss contributions over tasks. Empirical results across CIFAR-10/100, ImageNet100, and GNSS interference datasets show BLCL achieving superior average accuracies and better cluster structure than strong baselines such as Replay, MEMO, and DER, with interpretable improvements in confusion matrices and embedding visualizations. This approach provides a scalable, uncertainty-aware framework for continual learning with robust representation learning in diverse domains.

Abstract

The primary objective of methods in continual learning is to learn tasks in a sequential manner over time (sometimes from a stream of data), while mitigating the detrimental phenomenon of catastrophic forgetting. This paper proposes a method to learn an effective representation between previous and newly encountered class prototypes. We propose a prototypical network with a Bayesian learning-driven contrastive loss (BLCL), tailored specifically for class-incremental learning scenarios. We introduce a contrastive loss that incorporates novel classes into the latent representation by reducing intra-class and increasing inter-class distance. Our approach dynamically adapts the balance between the cross-entropy and contrastive loss functions with a Bayesian learning technique. Experimental results conducted on the CIFAR-10, CIFAR-100, and ImageNet100 datasets for image classification and images of a GNSS-based dataset for interference classification validate the efficacy of our method, showcasing its superiority over existing state-of-the-art approaches. Git: https://gitlab.cc-asp.fraunhofer.de/darcy_gnss/gnss_class_incremental_learning

Figures (136)

• Figure 1: Overview of our BLCL approach, which employs Bayesian weighting to balance cross-entropy and contrastive learning loss functions, facilitating efficient representations during updating specialized blocks.
• Figure 2: Overview of our BLCL method. After the generalized component of ResNet18 or ResNet32 he_zhang, follow up to eight blocks ($l = 8$) for the CIFAR krizhevsky datasets and two blocks ($l = 2$) for the GNSS ott_heublein_icl dataset in their respective specialized blocks. After each task the weights are averaged with the weights of the previous tasks. We propose a dynamic architecture with a different number of blocks and convolutional layers, indicated by $\text{BLCL}_{l[s]}$. After the FC layer, follows a CE loss in combination with a Bayesian learning-driven weighting of the contrastive loss (CL).
• Figure 3: 0.
• Figure 4: 1.
• Figure 5: 2.