Beyond filters: Compact feature map for portable deep model

Publication Type:
Conference Proceeding
Citation:
34th International Conference on Machine Learning, ICML 2017, 2017, 8 pp. 5672 - 5680
Issue Date:
2017-01-01
Metrics:
Full metadata record
Files in This Item:
Filename Description Size
wang17m.pdfPublished version307.82 kB
Adobe PDF
© Copyright 2017 by the authors(s). Convolutional neural networks (CNNs) have shown extraordinary performance in a number of applications, but they are usually of heavy design for the accuracy reason. Beyond compressing the filters in CNNs, this paper focuses on the redundancy in the feature maps derived from the large number of filters in a layer. We propose to extract intrinsic representation of the feature maps and preserve the discriminability of the features. Circulant matrix is employed to formulate the feature map transformation, which only requires O(dlogd) computation complexity to embed a d-dimensional feature map. The filter is then reconfigured to establish the mapping from original input to the new compact feature map, and the resulting network can preserve intrinsic information of the original network with significantly fewer parameters, which not only decreases the online memory for launching CNN but also accelerates the computation speed. Experiments on benchmark image datasets demonstrate the superiority of the proposed algorithm over state-of-the-art methods.
Please use this identifier to cite or link to this item: