Joint distribution optimal transportation for domain adaptation
Authors: Nicolas Courty, Rémi Flamary, Amaury Habrard, Alain Rakotomamonjy
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We finally provide several numerical experiments on real regression and classification problems that show the performances of JDOT over the state-of-the-art (Section 5). |
| Researcher Affiliation | Academia | Nicolas Courty Université de Bretagne Sud, IRISA, UMR 6074, CNRS, courty@univ-ubs.fr Rémi Flamary Université Côte d Azur, Lagrange, UMR 7293 , CNRS, OCA remi.flamary@unice.fr Amaury Habrard Univ Lyon, UJM-Saint-Etienne, CNRS, Lab. Hubert Curien UMR 5516, F-42023 amaury.habrard@univ-st-etienne.fr Alain Rakotomamonjy Normandie Universite Université de Rouen, LITIS EA 4108 alain.rakoto@insa-rouen.fr |
| Pseudocode | Yes | This algorithm well-known as Block Coordinate Descent (BCD) or Gauss-Seidel method (the pseudo code of the algorithm is given in appendix). |
| Open Source Code | Yes | Open Source Python implementation of JDOT: https://github.com/rflamary/JDOT |
| Open Datasets | Yes | Caltech-Office classification dataset. This dataset [28] is dedicated to visual adaptation. It contains images from four different domains: Amazon, the Caltech-256 image collection, Webcam and DSLR. |
| Dataset Splits | Yes | All the methods have hyper-parameters, that are selected using the reverse cross-validation of Zhong and colleagues [31].The dimension d for SA is chosen from {1, 4, 7, . . . , 31}. The entropy regularization for OT-IT and OT-MM is taken from {102, . . . , 105}, 102 being the minimum value for the Sinkhorn algorithm to prevent numerical errors. Finally the η parameter of OT-MM is selected from {1, . . . , 105} and the α in JDOT from {10 5, 10 4, . . . , 1}. |
| Hardware Specification | No | The paper does not provide specific details on the hardware used for experiments, such as CPU or GPU models, or memory specifications. |
| Software Dependencies | No | The paper mentions 'Open Source Python implementation of JDOT' and discusses using SVMs and neural networks, but it does not specify version numbers for Python, any libraries, or other software components used for reproducibility. |
| Experiment Setup | Yes | The dimension d for SA is chosen from {1, 4, 7, . . . , 31}. The entropy regularization for OT-IT and OT-MM is taken from {102, . . . , 105}, 102 being the minimum value for the Sinkhorn algorithm to prevent numerical errors. Finally the η parameter of OT-MM is selected from {1, . . . , 105} and the α in JDOT from {10 5, 10 4, . . . , 1}. The neural network used for all methods in this experiment is a simple 2-layer model with sigmoid activation function in the hidden layer to promote non-linearity. 50 neurons are used in this hidden layer. For DANN, hyper-parameters are set through the reverse cross-validation proposed in [11], and following the recommendation of authors the learning rate is set to 10 3. In the case of JDOT, we used the heuristic setting of α = 1/ maxi,j d(xs i, xt j)... 10 iterations of the block coordinate descent are realized. For each method, we stop the learning process of the network after 5 epochs. |