Multi-task Data Mining Based on Dynamic Representation Bias

点击翻译按钮获取中文摘要

To effectively cope with multiple pattern discovery tasks related each other, we have developed novel data mining methods each of which automatically modifies representations of data and patterns, implemented them as computer systems, and demonstrated their effectiveness with synthetic and real data. Remarkable achievements are multi-task classification method which employs an extended MDL principle to allow a common dictionary, a multi-task clustering method which employs an extension of information distance based on Kolmogorov complexity, and a dimension reduction method for multi-task data mining.