Semi-Supervised Multi-Task Learning Based On DFS

Recently semi-supervised multi-task learning has been one of the hotest problems in machine learning. As is known to us all, we can find the ubiquity of the dynamic fuzzy problems easily, and dynamic fuzzy set, i.e. DFS, is an effective methodology to solve the dynamic fuzzy problems. Therefore, in this paper we propose a new semi-supervised multi-task learning model based on DFS, which tries to integrate the semi-supervised learning and multi-task learning into one framework by using the DFS theory. Our main work is as follws:(1) The current stituation of semi-supervised learning, multi-task learning and semi-supervised multi-task learning has been analysed.(2) A new semi-supervised multi-task learning framework based on DFS theory has been proposed.(3) Two algorithms, called dynamic fuzzy semi-supervised multi-task pattern matching algorithm and dynamic fuzzy semi-supervised multi-task adaptive learning algorithm respectively, have been presented. Experiments show that both of them are excutable and effective.(4) Finally we apply our model into the face recognition application, and the results represent that the model proposed has a good advantage.In a word, the semi-supervised multi-task learning has been enriched with our new model. On the other hand, it’s also a new theoretical method provided for semi-supervised multi-task learning. Although good results have been achieved, there’s still some work left to be studied, such as the algorithm model optimizaiton problem, how to determine the algorithm parameters automatically rather than manually and so on.