Invertible Residual Networks
This addresses the need for versatile neural network architectures that can handle multiple tasks without architectural restrictions, though it is incremental in combining existing concepts.
The paper tackled the problem of making standard ResNet architectures invertible to enable a single model for classification, density estimation, and generation, achieving competitive performance with state-of-the-art image classifiers and flow-based generative models.
We show that standard ResNet architectures can be made invertible, allowing the same model to be used for classification, density estimation, and generation. Typically, enforcing invertibility requires partitioning dimensions or restricting network architectures. In contrast, our approach only requires adding a simple normalization step during training, already available in standard frameworks. Invertible ResNets define a generative model which can be trained by maximum likelihood on unlabeled data. To compute likelihoods, we introduce a tractable approximation to the Jacobian log-determinant of a residual block. Our empirical evaluation shows that invertible ResNets perform competitively with both state-of-the-art image classifiers and flow-based generative models, something that has not been previously achieved with a single architecture.