Taylan Cemgil, Research Scientist, DeepMind UK will speak at the Jean Golding Institute Data Seminar Series and an exclusive talk specifically for Compass students on Representation Learning.
Speaking at the Jean Golding Institute Data Seminar Series, Taylan Cemgil, Research Scientist, DeepMind UK will give a talk on:
‘Machine learning systems are not robust by default. Even systems that are reported to outperform humans in a particular domain can be often shown to fail at solving problems with virtually small variations on the problem data. This talk will focus on robustness in unsupervised learning and representation learning. In particular, we will give an outline of the current work on robust training. Our goal will be to highlight the nature of the challenges that are faced in ensuring that learning systems work according to desired specifications.’
‘Machine learning systems are not robust by default. Even systems that are reported to outperform humans in a particular domain can be often shown to fail at solving problems with virtually small variations on the problem data. This talk will focus on robustness in unsupervised learning and representation learning. In particular, we will give an outline of the current work on robust training. Our goal will be to highlight the nature of the challenges that are faced in ensuring that learning systems work according to desired specifications.’
In addition to the talk, and specifically for Compass students, he will give a Tutorial on Representation Learning:
Abstract: Representation learning is a fundamental problem in Machine learning and holds the promise to enable data-efficient learning and transfer to new tasks. In many domains such as computer vision or natural language processing, researchers have demonstrated the effectiveness of representations and features computed by deep architectures for the solution of other tasks. In this talk we will introduce basic ideas in representation learning, with a focus on approaches based on Variational Autoencoders, a model that is based on a decoder and encoder.