Hartley Spectral Pooling for Deep Learning
This work addresses the need for efficient and less lossy pooling methods in CNNs, though it is incremental as it builds on existing spectral pooling approaches.
The authors tackled the problem of lossy dimensionality reduction in CNNs by proposing Hartley spectral pooling, which reduces computational complexity compared to Fourier spectral pooling and preserves more structural features than max/average pooling, leading to improved convergence on MNIST and CIFAR-10 datasets.
In most convolution neural networks (CNNs), downsampling hidden layers is adopted for increasing computation efficiency and the receptive field size. Such operation is commonly so-called pooling. Maximation and averaging over sliding windows (max/average pooling), and plain downsampling in the form of strided convolution are popular pooling methods. Since the pooling is a lossy procedure, a motivation of our work is to design a new pooling approach for less lossy in the dimensionality reduction. Inspired by the Fourier spectral pooling(FSP) proposed by Rippel et. al. [1], we present the Hartley transform based spectral pooling method in CNNs. Compared with FSP, the proposed spectral pooling avoids the use of complex arithmetic for frequency representation and reduces the computation. Spectral pooling preserves more structure features for network's discriminability than max and average pooling. We empirically show that Hartley spectral pooling gives rise to the convergence of training CNNs on MNIST and CIFAR-10 datasets.