Self Normalizing Flows

Authors: Thomas A Keller, Jorn W.T. Peters, Priyank Jaini, Emiel Hoogeboom, Patrick Forré, Max Welling

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show experimentally that such models are remarkably stable and optimize to similar data likelihood values as their exact gradient counterparts, while training more quickly and surpassing the performance of functionally constrained counterparts.
Researcher Affiliation Collaboration 1Uv A-Bosch Delta Lab 2University of Amsterdam, Netherlands.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes We provide extended explanations for these discrepancies, as well as a link to our code repository, in the appendix (See Section A.3).
Open Datasets Yes In this framework we train models on MNIST, CIFAR-10, and the downsized Imagenet 32x32 dataset.
Dataset Splits Yes Figure 5. Negative log-likelihood on the MNIST validation set for a 2-layer fully connected flow trained with exact vs. self normalizing (SNF) gradients.
Hardware Specification No Table 3. Runtime comparison for the models presented in Tables 1 and 2. Hardware and implementation details are in Section A.3. (The main text defers hardware details to the appendix and does not provide specific models or specifications).
Software Dependencies No Table 3. Runtime comparison for the models presented in Tables 1 and 2. Hardware and implementation details are in Section A.3. (The main text defers software dependency details to the appendix and does not provide specific versions).
Experiment Setup No All training details can be found in the appendix (see Section A.3). (The main text defers explicit experimental setup details like hyperparameters and training configurations to the appendix).