An SVM-based incremental learning algorithm for user adaptation of sketch recognition

User adaptation is a critical problem in the design of human-computer interaction systems. Many pattern recognition problems, such as handwriting/sketching recognition and speech recognition, are user dependent, since different users' handwritings, drawing styles, and accents are different. Therefore, the classifiers for these problems should provide the functionality of user adaptation so as to let each particular user experience better recognition accuracy according to his input habit/style. However, the user adaptation functionality requires the classifiers to have the incremental learning ability, by which the classifiers can adapt to the user quickly without too much computation cost. In this paper, an SVM-based incremental learning algorithm is presented to solve this problem for sketch recognition. Our algorithm utilizes only the support vectors instead of all the historical samples, and selects some important samples from all newly added samples as training data. The importance of a sample is measured according to its distance to the hyper-plane of the SVM classifier. Theoretical analysis, experimentation, and evaluation of our algorithm in our online graphics recognition system SmartSketchpad, are presented to show the effectiveness of this algorithm. According to our experiments, this algorithm can reduce both the training time and the required storage space for the training dataset to a large extent with very little loss of precision.