Fixed budget quantized kernel least-mean-square algorithm

We present a quantization-based kernel least mean square (QKLMS) algorithm with a fixed memory budget. In order to deal with the growing support inherent in online kernel methods, the proposed method utilizes a growing and pruning combined technique and defines a criterion, significance, based on weighted statistical contribution of a data center. This method doesn’t need any apriori information and its computational complexity is acceptable, linear with the center number. As we show theoretically and experimentally, the introduced algorithm successfully quantifies the least ‘significant’ datum and preserves the most important ones resulting in less system error.