Knowledge Transfer using Multiresolution Learning

For my dissertation research, I propose to explore the transfer of knowledge at multiple levels of abstraction to improve learning. These multiple levels of abstraction will be created using multiresolution analysis, providing a principled and formal mechanism for abstracting knowledge. I claim that by exploiting the similarities between objects at various levels of detail, learning at multiple resolutions can facilitate transfer between related tasks.

The use of multiple resolutions allows the selective transfer of knowledge at specific levels of generalization between tasks. The proposed work focuses on two mechanisms for performing multiresolution transfer. The first method, data-based multiresolution transfer, uses multiple resolutions of input data to create models at different resolutions. The second method, model-based multiresolution transfer, generates multiple resolutions of previously learned models and then selectively transfers the appropriate resolution of the model. An additional contribution of this work will be a general framework for knowledge transfer that provides a foundation for comparing different transfer methods.