Post Selection Inference with Incomplete Maximum Mean Discrepancy Estimator
Authors: Makoto Yamada, Denny Wu, Yao-Hung Hubert Tsai, Hirofumi Ohta, Ruslan Salakhutdinov, Ichiro Takeuchi, Kenji Fukumizu
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through synthetic and real-world feature selection experiments, we show that the proposed framework can successfully detect statistically significant features. Last, we propose a sample selection framework for analyzing different members in the Generative Adversarial Networks (GANs) family. |
| Researcher Affiliation | Academia | Makoto Yamada1,2,3,4 , Denny Wu5,6 , Yao-Hung Hubert Tsai7, Hirofumi Ohta8, Ichiro Takeuchi9, Ruslan Salakhutdinov7, Kenji Fukumizu2,4 Kyoto University1, RIKEN AIP2, JST PRESTO3, Institute of Statistical Mathematics4, University of Toronto5, Vector Institute6, Carnegie Mellon University7, University of Tokyo8, Nagoya Institute of Technology9 |
| Pseudocode | Yes | Algorithm 1 mmd Inf (Feature Selection) |
| Open Source Code | No | The paper does not explicitly state that its own source code for the proposed methodology is publicly available, nor does it provide a direct link to it. It only references a third-party GAN package (Chainer GAN package) used in their experiments. |
| Open Datasets | Yes | generated 5000 images (using Chainer GAN package 1 with CIFAR10 datasets) |
| Dataset Splits | No | The paper does not provide specific train/validation/test dataset splits, such as exact percentages or sample counts for each partition. It mentions using '1/2 of data to calculate the covariance matrix of MMD and the rest to perform feature selection and inference' but this is not a standard model validation split. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions using 'Chainer GAN package' and 'pre-trained Resnet18' but does not specify version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | We fixed the number of selected features (prior to PSI) k to 30. ... features with p-value lower than the significance level α = 0.05 are selected as statistically significant features. For block MMD, in each experiment we set the candidate of block size as B = {10, 20, 50}. For incomplete MMD, in each experiment the ratio between number of pairs (i, j) sampled to compute incomplete MMD score and sample size is fixed at r = ℓ n {0.5, 5, 10}. |