Auto-Encoding Goodness of Fit
Authors: Aaron Palmer, Zhiyi Chi, Derek Aguiar, Jinbo Bi
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, we show that our higher criticism parameter selection procedure balances reconstruction and generation using mutual information and uniformity of p-values respectively. Finally, we show that Go FAE achieves comparable FID scores and mean squared errors with competing deep generative models while retaining statistical indistinguishability from Gaussian in the latent space based on a variety of hypothesis tests. We evaluate generation and reconstruction performance, normality, and informativeness of Go FAE1 using several Go F statistics on the MNIST (Le Cun et al., 1998), Celeb A (Liu et al., 2015) and CIFAR10 (Krizhevsky et al., 2009) datasets and compare to several other GAE models. |
| Researcher Affiliation | Academia | Aaron Palmer1, Zhiyi Chi2, Derek Aguiar1, Jinbo Bi1 1Department of Computer Science, 2Department of Statistics {aaron.palmer,zhiyi.chi,derek.aguiar,jinbo.bi}@uconn.edu University of Connecticut, Storrs, CT, USA |
| Pseudocode | Yes | Algorithm 1 Evaluating Higher Criticism (Section 3.2), Algorithm 2 Go FAE Optimization (Section 5), Algorithm S1 Go FAE Pipeline: Training Locally and Assessing with Higher Criticism (Appendix D.4). |
| Open Source Code | Yes | Code can be found at https://github.com/aripalmer/Go FAE. |
| Open Datasets | Yes | We evaluate generation and reconstruction performance, normality, and informativeness of Go FAE1 using several Go F statistics on the MNIST (Le Cun et al., 1998), Celeb A (Liu et al., 2015) and CIFAR10 (Krizhevsky et al., 2009) datasets and compare to several other GAE models. |
| Dataset Splits | Yes | We selected λ with grid-search using a training and validation set and an array of λ values. After setting a seed, the training set was split into a smaller training set, of 45 thousand, and a validation set with the rest. |
| Hardware Specification | No | No specific hardware details (e.g., GPU/CPU models, cloud resources) used for running experiments were mentioned in the paper. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify its version number or versions for other software dependencies. |
| Experiment Setup | Yes | Training proceeded for 50 epochs using mini-batches of size 128. We specified a 10-dimensional code layer. For our models, Adam optimizer Kingma & Ba (2014) was used for the encoder (learning rate 3e 3, β = (0.5, 0.999)) and decoder (learning rate 3e 3, β = (0.5, 0.999)). Riemannian SGD with learning rate 5e 3 was used to constrain Θ to the Stiefel manifold using a one cycle learning rate scheduler with max learning rate as 1e 3. |