Learning Discriminatively Reconstructed Source Data for Object Recognition with Few Example

0
771
Learning Discriminatively Reconstructed Source Data for Object Recognition with Few Example

Learning Discriminatively Reconstructed Source Data for Object Recognition with Few Example

Abstract of Learning Discriminatively Reconstructed Source Data for Object

Learning Discriminatively Reconstructed Source Data for Object

Learning Discriminatively Reconstructed Source Data for Object Recognition with Few Example,We aim at improving the object recognition with few training data in the target domain by leveraging abundant auxiliary data in the source domain. The major issue obstructing knowledge transfer from source to target is the limited correlation between the two domains. Transferring irrelevant information from the source domain usually leads to performance degradation in the target domain. To address this issue, we propose a transfer learning framework with the two key components, such as discriminative source data reconstruction and dual-domain boosting. The former correlates the two domains via reconstructing source data by target data in a discriminative manner.

Conclusions

Learning Discriminatively Reconstructed Source Data for Object Recognition with Few Example,We have presented an effective approach that can leverage useful knowledge in the source domain to facilitate classifier learning in the target domain, especially when few training examples in target are available. Our approach makes no assumption about the correlation between the source and target domains, since it can correlate the two domains via reconstructing source data by target data in a discriminative manner. During the process of reconstruction, the source data are firstly transformed to adapt inter-domain variations, and only the reconstructed part is borrowed to enrich the corresponding training set in target. The developed dual-domain boosting then casts knowledge transfer as a multi-task learning problem.