Invertible Convolutional Flow

Authors: Mahdi Karami, Dale Schuurmans, Jascha Sohl-Dickstein, Laurent Dinh, Daniel Duckworth

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We first conduct experiments to evaluate the benefits of the proposed flow model (CONF). We compare the density estimation performance of CONF to the affine coupling flow models real NVP [Dinh et al., 2016] and Glow [Kingma and Dhariwal, 2018], and the recent continuous-time invertible generative model FFJORD [Grathwohl et al., 2019].
Researcher Affiliation Collaboration Mahdi Karami Department of Computer Science University of Alberta karami1@ualberta.ca Jascha Sohl-Dickstein Dale Schuurmans Laurent Dinh Daniel Duckworth Google Brain
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any explicit statement or link for open-source code for the methodology described.
Open Datasets Yes We also perform unconditional density estimation on two image datasets; MNIST, consisting of handwritten digits [Y. Le Cun, 1998] and CIFAR-10, consisting of natural images [Krizhevsky, 2009].
Dataset Splits No Details of model architecture and experimental setup together with more empirical results are presented in appendix. The paper refers to standard datasets and implicitly uses test sets (Table 1), but explicit train/validation/test splits (e.g., percentages or counts) are not provided in the main text.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes Each coupling flow is composed of a maximum of M = 2 iterates of the combined convolution flow. the parameters of the nonlinear bijector pair, {σα, σα }, are initialized sufficiently close to zero so that they behave approximately as linear functions at the outset. Furthermore, the conditioning networks are initialized such that the scaling filters, s, and the convolution kernels at the frequency domain, F{w}, are all initially identity filters.