Structured Receptive Fields in CNNs

TL;DR

Structured receptive field networks combine fixed basis functions with learnable filter sets, improving CNN performance on small datasets and in domains lacking large pre-training datasets, bridging classical analysis and modern CNNs.

Contribution

The paper introduces structured receptive field networks that use a fixed basis with learnable weights, enhancing CNN flexibility and performance especially on limited data.

Findings

Significant improvement over unstructured CNNs on small and medium datasets.

Outperforms scattering networks on large datasets.

Achieves state-of-the-art results on 3D MRI brain-disease classification.

Abstract

Learning powerful feature representations with CNNs is hard when training data are limited. Pre-training is one way to overcome this, but it requires large datasets sufficiently similar to the target domain. Another option is to design priors into the model, which can range from tuned hyperparameters to fully engineered representations like Scattering Networks. We combine these ideas into structured receptive field networks, a model which has a fixed filter basis and yet retains the flexibility of CNNs. This flexibility is achieved by expressing receptive fields in CNNs as a weighted sum over a fixed basis which is similar in spirit to Scattering Networks. The key difference is that we learn arbitrary effective filter sets from the basis rather than modeling the filters. This approach explicitly connects classical multiscale image analysis with general CNNs. With structured receptive field networks, we improve considerably over unstructured CNNs for small and medium dataset scenarios as well as over Scattering for large datasets. We validate our findings on ILSVRC2012, Cifar-10, Cifar-100 and MNIST. As a realistic small dataset example, we show state-of-the-art classification results on popular 3D MRI brain-disease datasets where pre-training is difficult due to a lack of large public datasets in a similar domain.