Regularizing CNNs with Locally Constrained Decorrelations

TL;DR

This paper introduces OrthoReg, a novel regularization method that enforces local feature orthogonality in CNNs, effectively reducing overfitting and improving accuracy, especially in fully convolutional networks.

Contribution

OrthoReg is a new regularization technique that locally constrains feature decorrelation by enforcing orthogonality on weights, enhancing model capacity utilization.

Findings

OrthoReg improves accuracy bounds even with batch normalization and dropout.

It effectively reduces overfitting on CIFAR-10, CIFAR-100, and SVHN datasets.

The method is particularly suitable for fully convolutional neural networks.

Abstract

Regularization is key for deep learning since it allows training more complex models while keeping lower levels of overfitting. However, the most prevalent regularizations do not leverage all the capacity of the models since they rely on reducing the effective number of parameters. Feature decorrelation is an alternative for using the full capacity of the models but the overfitting reduction margins are too narrow given the overhead it introduces. In this paper, we show that regularizing negatively correlated features is an obstacle for effective decorrelation and present OrthoReg, a novel regularization technique that locally enforces feature orthogonality. As a result, imposing locality constraints in feature decorrelation removes interferences between negatively correlated feature weights, allowing the regularizer to reach higher decorrelation bounds, and reducing the overfitting more effectively. In particular, we show that the models regularized with OrthoReg have higher accuracy bounds even when batch normalization and dropout are present. Moreover, since our regularization is directly performed on the weights, it is especially suitable for fully convolutional neural networks, where the weight space is constant compared to the feature map space. As a result, we are able to reduce the overfitting of state-of-the-art CNNs on CIFAR-10, CIFAR-100, and SVHN.