下星期一(12月28日)下午1點半到3點於EC329教室

我很高興能夠邀請到研究表傑非常傑出的Mr. Chin-Wei Huang, PhD student from University of Montreal來給予演講
"Flows and Friends: A Unifying Perspective of Likelihood-Based Generative Models"

Flow model是近年來開始受到矚目並開始被大量使用的deep models

摘要：

In this talk, I will give a short introduction to invertible neural networks used as flows of probability distributions. I will then present some recent works on improving the expressiveness
of invertible flows by means of universal monotonic neural networks, variational augmentation, and input-convex neural networks. This talk will examine the broader family of likelihood-based generative models through the lens of reparameterization -- unifying variational autoencoders, autoregressive models and normalizing flows.

講者Bio：

Chin-Wei is a PhD candidate at the University of Montreal, affiliated with Mila, the Quebec Artificial Intelligence Institute, advised by Prof Aaron Courville. His research lies at the intersection of probabilistic modelling and deep learning, with a focus on improving the representational power of invertible flow-based models for modelling probability distributions. During his PhD, Chin-Wei has also spent some time interning at Element AI and Google. He is a Google PhD fellow in Machine Learning as of 2020.