Fighting Our Big Data Addiction With Representation Learning
Recently, tremendous progress has been made in applying machine learning to AI problems (e.g. speech recognition, object recognition, machine translation), notably with deep neural networks, that automate the learning of meaningful representations (features) directly from raw data. However, these successes were largely enabled thanks to the collection of large labeled datasets. Such datasets are costly to generate, thus it would be desirable to reduce our reliance on them.
In this talk, I'll discuss 3 general approaches that go in this direction: learning representations using unsupervised learning, multi-modal learning and domain adaptation. For each, I'll give an example of a successful instantiation of these approaches, taken from my own research.
Hugo Larochelle is Research Scientist at Twitter and Assistant Professor at the Université de Sherbrooke (UdeS). Before, he spent two years in the machine learning group at University of Toronto, as a postdoctoral fellow under the supervision of Geoffrey Hinton. He obtained his Ph.D. at Université de Montréal, under the supervision of Yoshua Bengio. He is the recipient of two Google Faculty Awards. His professional involvement includes associate editor for the IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), member of the editorial board of the Journal of Artificial Intelligence Research (JAIR) and program chair for the International Conference on Learning Representations (ICLR) of 2015, 2016 and 2017.