In this paper, we provide a solution for the biggest and most significant challenge for learning from streaming data, i.e., scarcity of labeled data. The proposed algorithm exploits the online micro-clusters to summarize the streaming data in a compact form and are further used to classify the incoming data stream instances. Our algorithm provides a reliability model by learning the importance of these micro-clusters over time by error-driven representative learning. The main advantage of our algorithm is that it is light-weight and work under any computational resources, i.e., it works under limited memory and learns concept-drifting streaming data in online fashion.
This is the version 1, and it will be constantly improved. We will update the progress.
Step 1. input "data.mat"; Step 2. run Main.m.
Reference: Salah Ud Din, Junming Shao, Jay Kumar, Waqar Ali, Jiaming Liu, Yu Ye, Online Reliable Semi-supervised Learning on Evolving Data Streams, Information Sciences, 2020, https://doi.org/10.1016/j.ins.2020.03.052.
ATTN: This code were developed by Salah Ud Din (salahuddin@std.uestc.edu.cn). For any problem and suggestment, please feel free to contact Mr. Salah Ud Din.