Simone Bonechi (DIISM, University of Siena)
Nov 21, 2018 – 11:00 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
In recent years, Deep Neural Networks (DNNs) led to impressive results in a wide variety of machine learning tasks, tipically relying on the existence of a huge amount of supervised data. However, in many applications, such as bio–medical image analysis, gathering large sets of labeled data could be very difficult and costly.
Unsupervised domain adaptation exploits data from a source domain, where annotations are available, to train a model capable to generalize also to a target domain, where labels are scarce or unavailable.
Recent research has shown that Generative Adversarial Networks (GANs) can be successfully employed for domain adaptation, although deciding when to stop learning is a major problem for GANs.
In this seminar, I propose some confidence measures that can be used to early stop the GAN training, also showing how such confidence measures can be employed to predict the reliability of the network output.
The effectiveness of the proposed measures were tested in two domain adaptation tasks, with very promising results.