Recursive Autoconvolution for Unsupervised Learning of Convolutional Neural Networks
This addresses the problem of efficient visual recognition using cheap unlabeled data, though it appears incremental as it builds on existing unsupervised methods and network designs.
The paper tackles unsupervised learning for image classification by applying the recursive autoconvolution operator to boost discriminative filter learning, achieving state-of-the-art results on benchmarks like MNIST, CIFAR-10, CIFAR-100, and STL-10.
In visual recognition tasks, such as image classification, unsupervised learning exploits cheap unlabeled data and can help to solve these tasks more efficiently. We show that the recursive autoconvolution operator, adopted from physics, boosts existing unsupervised methods by learning more discriminative filters. We take well established convolutional neural networks and train their filters layer-wise. In addition, based on previous works we design a network which extracts more than 600k features per sample, but with the total number of trainable parameters greatly reduced by introducing shared filters in higher layers. We evaluate our networks on the MNIST, CIFAR-10, CIFAR-100 and STL-10 image classification benchmarks and report several state of the art results among other unsupervised methods.