Noise robust neural network architecture
This work addresses noise robustness in image recognition for applications like computer vision, but it is incremental as it builds on existing neural network methods with specific modifications.
The authors tackled the problem of recognizing noisy images without augmenting training data by proposing a dune neural network that represents parameters as uncertainty intervals and applies linear transformations to inputs, achieving better test accuracy than humans on very noisy MNIST images.
In which we propose neural network architecture (dune neural network) for recognizing general noisy image without adding any artificial noise in the training data. By representing each free parameter of the network as an uncertainty interval, and applying a linear transformation to each input element, we show that the resulting architecture achieves decent noise robustness when faced with input data with white noise. We apply simple dune neural networks for MNIST dataset and demonstrate that even for very noisy input images which are hard for human to recognize, our approach achieved better test set accuracy than human without dataset augmentation. We also find that our method is robust for many other examples with various background patterns added.