A Swiss Army Knife for Minimax Optimal Transport
Authors: Sofien Dhouib, Ievgen Redko, Tanguy Kerdoncuff, Rémi Emonet, Marc Sebban
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we provide an experimental study highlighting the efficiency of our approach. |
| Researcher Affiliation | Academia | 1Univ Lyon, INSA-Lyon, Universit e Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR 5220, U1206, F-69100, LYON, France 2Univ Lyon, UJM-Saint-Etienne, CNRS, Institut d Optique Graduate School Laboratoire Hubert Curien UMR 5516, F-42023, Saint-Etienne, France. |
| Pseudocode | Yes | We further use the constraint dropping strategy (Mutapcic & Boyd, 2009, Sec. 5.3.2) and provide a complete pseudo-code for our algorithm in Algorithm 1, where thd1 and thd2 respectively control the stopping criterion and the constraint elimination. |
| Open Source Code | Yes | The code for the different experiments is available on this link3. https://github.com/sofiendhouib/minimax_OT. |
| Open Datasets | Yes | The latter one is composed of 100 zeros and 100 ones coming from the MNIST dataset, after reducing its dimensionality to 10 with UMAP (Mc Innes et al., 2018). We consider the fragmented hypercube dataset studied in (Paty & Cuturi, 2019) and earlier in (Forrow et al., 2019). |
| Dataset Splits | No | The paper does not provide specific dataset split information (percentages, sample counts, or detailed splitting methodology) needed to reproduce data partitioning for training, validation, or testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Sinkhorn algorithm (Cuturi, 2013)' and 'UMAP (Mc Innes et al., 2018)' but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | The convergence of Algorithm 1 is illustrated in Figure 3 (left) by plotting the evolution of the quantity err(t) := |µt Pt, Ct | along the iterations for |C| {10, 40, 90}. We also test our algorithm with the entropic regularization of the transport matrix with λ {1, 0.1, 0.01} as regularization parameter... For a fixed C, let CC be defined as in (9)... thd1 and thd2 respectively control the stopping criterion and the constraint elimination. We consider only 200 pixels from each image. |