An Online Incremental Learning Vector Quantization

Abstract

As described in this paper, we propose online incremental learning vector quantization (ILVQ) for supervised classification tasks. As a prototype-based classifier, ILVQ needs no prior knowledge of the number of prototypes in the network or their initial value, as do most current prototype-based algorithms. It adopts a threshold-based insertion scheme to determine the number of prototypes needed for each class dynamically according to the distribution of training data. In addition, this insertion policy insures the fulfillment of the incremental learning goal, including both between-class incremental learning and within-class incremental learning. A technique for removing useless prototypes is used to eliminate noise interrupting the input data. Unlike other LVQ-based methods, the learning result won’t be affected by the sequence of input patterns that come into the ILVQ. Results of experiments described herein show that the proposed ILVQ can accommodate the non-stationary data environment and can provide good recognition performance and storage efficiency.KeywordsCompression RatioRecognition RateInput PatternVector QuantizationIncremental LearningThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.