Invertible DenseNets with Concatenated LipSwish
This work provides a more parameter-efficient and higher-performing invertible network architecture for researchers and practitioners working with flow-based models for density estimation and hybrid generative/discriminative tasks.
This paper introduces Invertible Dense Networks (i-DenseNets), which are more parameter-efficient than Residual Flows. The i-DenseNet architecture outperforms Residual Flow and other flow-based models on density estimation, measured in bits per dimension, given an equal parameter budget. It also outperforms Residual Flows when trained as a hybrid generative and discriminative model.
We introduce Invertible Dense Networks (i-DenseNets), a more parameter efficient extension of Residual Flows. The method relies on an analysis of the Lipschitz continuity of the concatenation in DenseNets, where we enforce invertibility of the network by satisfying the Lipschitz constant. Furthermore, we propose a learnable weighted concatenation, which not only improves the model performance but also indicates the importance of the concatenated weighted representation. Additionally, we introduce the Concatenated LipSwish as activation function, for which we show how to enforce the Lipschitz condition and which boosts performance. The new architecture, i-DenseNet, out-performs Residual Flow and other flow-based models on density estimation evaluated in bits per dimension, where we utilize an equal parameter budget. Moreover, we show that the proposed model out-performs Residual Flows when trained as a hybrid model where the model is both a generative and a discriminative model.