Adaptive sharing for image classification

摘要: In this paper, we formulate the image classification problem in a multi-task learning framework. We propose novel method to adaptively share information among tasks (classes). Different from imposing strong assumptions or discovering specific structures, key insight our is selectively extract and exploit shared classes while capturing respective disparities simultaneously. It achieved by estimating composite of two sets parameters with different regularization. Besides applying it for classifiers on pre-computed features, also integrate adaptive sharing deep neural networks, whose discriminative power can be augmented encoding class relationship. further develop strategies solving optimization problems scenarios. Empirical results demonstrate that significantly improve performance transferring knowledge appropriately.