| 研究実績の概要 |
Our project "onsite transfer learning (現地の転移学習)" finishes this year. In summary, all research activities under this grant has been conducted as it was planned, and most of the research targets have been met. Moreover, this research project has inspired us with new ideas that we believe are important and should be investigated in near future.
Specifically, We have 1) obtained an effective estimator for conditional density ratio, which can be solved efficiently without using expensive computational resources. 2) We have shown in our experiments, in many tasks, such conditional density ratio estimator can help us transfer a pre-trained classifier given limited number of samples. 3) This estimator is shown to be theoretically consistent under mild conditions.
|
