Multiple-Source-Adaptation-using-Variational-R-nyi-Bound-Optimization

This project contains code for our paper "Multiple-Source Adaptation using Variational Rényi Bound Optimization".

It contains pytorch implementation of Variational Inference model, using different loss functions:

1. Maximizing the ELBO.
2. Maximizing Rényi Lower Bound with positive alpha.
3. Minimizing Rényi Upper Bound with negative alpha (VRLU).
4. Using Rényi Upper-Lower Bounds combination as the loss function (VRS).

In addition, it contains implementation of our VRS-MSA model using both DC-programming and SGD. We used 3 datasets: MNIST, USPS and SVHN.