Margin-aware Adversarial Domain Adaptation with Optimal Transport
Authors: Sofien Dhouib, Ievgen Redko, Carole Lartizien
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | From the obtained theoretical results, we derive a novel algorithmic solution for domain adaptation that introduces a novel shallow OT-based adversarial approach and outperforms other OT-based DA baselines on several simulated and real-world classification tasks. Finally, in the last section, we evaluate our algorithm on a toy data set and on a benchmark real-world problem. |
| 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 | No | The paper describes the optimization procedure using block coordinate descent and L-BFGS but does not include a formal pseudocode block or algorithm listing. |
| Open Source Code | Yes | The code for the different experiments is available on this link2. 2https://github.com/sofiendhouib/MADAOT. |
| Open Datasets | Yes | We carry on our experiments on the moons data set used in (Courty et al., 2015). Below, we consider the famous Amazon product reviews dataset (Blitzer et al., 2007) related to the sentiment analysis task. |
| Dataset Splits | Yes | we use them as a validation set to select the best hyper-parameters (defined in Proposition 4) via a 5-fold cross-validation procedure for 10 values of δ ranging from 10 2 to 102, and 10 values for ζ from {10 6 to 10 2, both on a logarithm scale. For each task, we use predefined sets of 2000 instances of source and target data samples for training, and keep 4000 instances of the target domain for testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. |
| Software Dependencies | No | The paper does not specify the version numbers of any software dependencies, libraries, or frameworks used (e.g., Python, PyTorch, scikit-learn versions). |
| Experiment Setup | Yes | Hyper-parameter tuning is a longstanding problem in unsupervised domain adaptation... we use them as a validation set to select the best hyper-parameters (defined in Proposition 4) via a 5-fold cross-validation procedure for 10 values of δ ranging from 10 2 to 102, and 10 values for ζ from {10 6 to 10 2, both on a logarithm scale. This is a rather standard procedure in unsupervised domain adaptation used in several other papers on the subject (Courty et al., 2015; Bousmalis et al., 2016). We use this procedure for the first dataset. As for the real-world data set, we run all experiments by setting δ = 1 and ζ = 10 5. For a fixed transport matrix Γ, minimize over w. To this end, we use the L-BFGS quasi-Newton method. |