Partial Wasserstein Covering
Authors: Keisuke Kawano, Satoshi Koide, Keisuke Otaki7115-7123
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments In our experiments, we considered a scenario in which we wished to select some data in the application dataset Dapp for the PWC problem (i.e., Dapp = Dcand). For PW-*-LP and PW-sensitivity-Ctrans, we used IBM ILOG CPLEX 20.01 (IBM ILOG CPLEX Optimization Studio 2013) for the dual simplex algorithm, where the sensitivity analysis for LP is also available. For PW-*-ent, we computed the entropic regularized version of the partial Wasserstein divergence using PyTorch (Paszke et al. 2019) on a GPU. |
| Researcher Affiliation | Industry | Keisuke Kawano, Satoshi Koide, Keisuke Otaki Toyota Central R&D Labs., Inc. {kawano, koide, otaki}@mosk.tytlabs.co.jp |
| Pseudocode | Yes | Algorithm 1: Data selection for the PWC problem with sensitivity analysis (PW-sensitivity-LP) 1: Input: Dapp, Ddev, Dcand 2: Output: S 3: S {}, t 0 4: while |S| < K do 5: Calculate PW2(Dapp, S Ddev) , and obtain g (t) j from the sensitivity analysis. 6: j = arg minj [[Ncand]],s(j) / S g (t) j 7: S S {s(j )} 8: t t + 1 9: end while |
| Open Source Code | No | The paper does not contain any statements about releasing its source code, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We employ subsets of the MNIST dataset (Le Cun, Cortes, and Burges 2010)... We adopted two datasets, BDD100K (Yu et al. 2020) and KITTI (Object Detection Evaluation 2012) (Geiger et al. 2013) as the application and development datasets, respectively. |
| Dataset Splits | No | For the quantitative evaluation of the missing pattern findings, we herein consider a scenario in which a category (i.e., label) is less included in the development dataset than in the application dataset. We employ subsets of the MNIST dataset (Le Cun, Cortes, and Burges 2010), where the development dataset contains 0 labels at a rate of 0.5%, whereas all labels are included in the application data in equal ratios (i.e., 10%). We randomly sampled 500 images from the validation split for each Dapp and Ddev. |
| Hardware Specification | Yes | All experiments were conducted with an Intel Xeon Gold 6142 CPU and an NVIDIA TITAN RTX GPU. |
| Software Dependencies | Yes | For PW-*-LP and PW-sensitivity-Ctrans, we used IBM ILOG CPLEX 20.01 (IBM ILOG CPLEX Optimization Studio 2013) for the dual simplex algorithm... For PW-*-ent, we computed the entropic regularized version of the partial Wasserstein divergence using PyTorch (Paszke et al. 2019) on a GPU. |
| Experiment Setup | Yes | We set the coefficient for entropy regularization ϵ = 0.01 and terminated the Sinkhorn iterations when the divergence did not change by at least maxi,j Cij 10 12 compared with the previous step, or the number of iterations reached 50,000. |