The program will be available shortly. Please check back later.
Presenter: Atsuto Maki, Professor of Computer Science, KTH Royal Institute of Technology, Sweden.
Deep convolutional neural networks have widely been applied to various tasks of image recognition. Despite fast improvements in handling training data, network designs, and learning algorithms, the full principle is yet to be explained; there are unique challenges due to the depth as well as some pitfalls in the behaviour of discriminative models.
In this talk, we will address some important but often overlooked problems in training a network from the viewpoint of generalization ability and reliability of output, and discuss some ways for alleviating them with respect to classification problems.
In this talk, we will address some important but often overlooked problems in training a network from the viewpoint of generalization ability and reliability of output, and discuss some ways for alleviating them with respect to classification problems.
In this talk, we will address some important but often overlooked problems in training a network from the viewpoint of generalization ability and reliability of output, and discuss some ways for alleviating them with respect to classification problems.