We propose ways to improve the performance of fully connected networks. We found that two approaches in particular have a strong effect on performance: linear bottleneck layers and unsupervised pre-training using autoencoders without hidden unit biases. We show how both approaches can be related to improving gradient flow and reducing sparsity in the network. We show that a fully connected network can yield approximately 70% classification accuracy on the permutation-invariant CIFAR-10 task, which is much higher than the current state-of-the-art. By adding deformations to the training data, the fully connected network achieves 78% accuracy, which is just 10% short of a decent convolutional network.
@article{arxiv.1511.02580,
title = {How far can we go without convolution: Improving fully-connected networks},
author = {Zhouhan Lin and Roland Memisevic and Kishore Konda},
journal= {arXiv preprint arXiv:1511.02580},
year = {2015}
}