Class-Incremental Quad-Channel Contrastive Prototype Networks (CI-QCCPN)
Memory replays for improving Open-Set Recognition (OSR) and Continual Learning (CL)
Open-Set Recognition (OSR) and Continual Learning (CL) have recently gained popularity from researchers studying to advance sustainable Artificial Intelligence (AI). However, there is less research focus on the performance of rejecting unknown classes in continual OSR models. Empirical evidence that continual OSR models tend to experience a decline in their ability to reject unknown classes over successive task periods has been provided.
To address this issue, Alfarisy, Malik and Ong propose a solution of using two memory replays: one for samples of known features and another for pseudo-features associated with unknown classes. For generating features of these unknown classes, a method called Perceptive Unknown Feature Search (PUFS) that involves locating features based on the positions of existing prototypes and then feeding them into an inverse network to obtain backbone features was proposed.
The loss function by incorporating contrastive learning for unknown features is modified to enhance the model. Named as Class-Incremental Quad-Channel
Contrastive Prototype Networks (CI-QCCPN), this improved model outperforms its predecessor QCCPN as well as softmax-based classifiers with memory replay and achieves the highest average AUROC scores across various tasks and datasets.
This work has been presented by Gusti Alfarisy at the The 6th International Conference on Applied Computational Intelligence in Information Systems in Brunei and have been awarded as best paper.