Główne pojęcia
Proposing a continual feature selection framework based on granular-ball knowledge transfer for open environments.
Streszczenie
The paper introduces a novel framework for continual feature selection (CFS) in data preprocessing, focusing on an open and dynamic environment. The CFS method combines continual learning with granular-ball computing to detect unknown classes and facilitate knowledge transfer. It consists of two stages: initial learning and open learning. The proposed method aims to establish an initial knowledge base through multi-granularity representation using granular-balls and then utilize prior knowledge to identify unknowns, update the knowledge base, reinforce old knowledge, and integrate new knowledge. Experimental results demonstrate the superiority of the method compared to state-of-the-art feature selection methods.
Statystyki
Xuemei Cao, Xin Yang*, Member, IEEE, Shuyin Xia, Member, IEEE, Guoyin Wang, Senior Member, IEEE, Tianrui Li, Senior Member, IEEE
Extensive experimental results on public benchmark datasets demonstrate the method's superiority in terms of effectiveness and efficiency compared to state-of-the-art feature selection methods.
For privacy protection reasons, historical data may not be accessible once hidden.
Granular-ball computing (GBC) focuses on constructing an optimal set of granular balls.
CL aims to facilitate knowledge transfer across a sequence of tasks.
Cytaty
"To mitigate this [catastrophic forgetting], a new learning paradigm, Continual Learning (CL), has been recently proposed."
"Granular-ball computing (GBC) is a new sample space representation method that is efficient, robust, and has good knowledge representation capabilities."
"The main contribution of this paper can be summarized as follows..."