Understanding the Limitations of Conditional Generative Models

Authors: Ethan Fetaya, Joern-Henrik Jacobsen, Will Grathwohl, Richard Zemel

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, we find that while we are able to train robust models for MNIST, robustness completely breaks down on CIFAR10. We relate this failure to various undesirable model properties that can be traced to the maximum likelihood training objective.
Researcher Affiliation Academia Vector Institute and University of Toronto {ethanf, jjacobs,wgrathwohl, zemel}@cs.toronto.edu
Pseudocode No The paper does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. Methods are described in prose and mathematical equations.
Open Source Code No The paper does not provide any explicit statements about releasing source code for the described methodology or links to a code repository.
Open Datasets Yes Experimentally, we find that while we are able to train robust models for MNIST, robustness completely breaks down on CIFAR10. We also model MNIST and CIFAR10 as is typically done in the literature.
Dataset Splits No The paper mentions using a 'test set' for evaluation (e.g., 'over the test set'), but it does not specify the exact percentages or counts for training, validation, and test splits, nor does it reference a specific standard split that details these proportions.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions 'pytorch' as a framework but does not specify its version number or any other software dependencies with their respective versions.
Experiment Setup Yes The models are optimized using Adam, for 150 epochs. The initial learning rate is 1e 3, decayed by a factor of 10 every 60 epochs. For the reweighted optimization the objective is loss = log(p(x|y))/D log(p(y|x)) (6) where D is the data dimension (3x32x32 for CIFAR, 1x32x32 for MNIST).