Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Optimal Transport for Domain Adaptation through Gaussian Mixture Models

Created by
  • Haebom

Author

Eduardo Fernandes Montesuma, Fred Maurice Ngol e Mboula, Antoine Souloumiac

Outline

This paper addresses the problem that machine learning systems operate under the assumption that training and test data are sampled from fixed probability distributions, but in practice this assumption is rarely confirmed due to changes in data acquisition conditions. Therefore, we study how to perform unsupervised domain adaptation with minimal access to data with new conditions to learn a model that is robust to changes in data distribution. In particular, we analyze distribution changes by utilizing optimal transport, which allows mapping between domains, but to address the high computational cost of existing optimal transport methods, we explore optimal transport between Gaussian mixture models (GMMs). Using GMMs effectively reduces computational complexity, and we show that it is more efficient than existing shallow domain adaptation methods through nine benchmarks and a total of 85 adaptation tasks, and that it scales well with respect to the number of samples (n) and dimensionality (d).

Takeaways, Limitations

•
Takeaways:
◦
We present an unsupervised domain adaptation method that is more efficient than existing methods through optimal transportation using Gaussian mixture models (GMMs).
◦
We present a method that is scalable in terms of sample number and dimensionality.
◦
We verify the efficiency and performance of the method through experimental results using real datasets.
•
Limitations:
◦
Because it uses a Gaussian mixture model (GMM), its application may be limited to complex data distributions that are not well represented by GMM.
◦
The type and scope of the benchmark dataset may be limited. Experiments on more diverse datasets may be required.
◦
Further research is needed on the generalization performance of the method presented in this paper.
👍
Made with Slashpage