Incremental multiclass open-set audio recognition

Incremental learning aims to learn new classes if they emerge while maintaining the performance for previously known classes. It acquires useful information from incoming data update existing models. Open-set recognition, however, requires ability recognize examples and reject new/unknown There are two main challenges in this matter. First, class discovery: algorithm needs not only but it must also detect unknown Second, model extension: after identified, be updated. Focusing on matter, we introduce incremental open-set multiclass support vector machine algorithms that can classify seen/unseen classes, using increase current with without entirely retraining system. Comprehensive evaluations carried out both open set recognition learning. For adopt openness test examines effectiveness of a varying number known/unknown labels. learning, adapt single novel each phase Experimental results show promising proposed methods, compared some representative previous methods.