Prof. Ran He
Speech Title: "Variational Image Analysis under Limited Computational Resource"
Image data tend to be high-dimensional and large-scale. When given infinite computational resource, machine learning algorithms can generate exact results (prohibitively expensive). Variational approximation methods arise from the use of a finite amount of processor time. These methods are often built on top of standard function approximators. In this talk, we introduce a group of variational inference and learning algorithms that scale to high-dimensional and large-scale image datasets. First, we address the linear approximation to learn robust and compact local features of image data, named ordinal measures. Second, we address the quadratic approximation of a family of loss functions that widely used in image analysis. Accordingly, a half-quadratic optimization framework is proposed for modeling sparsity, low-rank recovery and noise. Third, we introduce an Introspective Variational Autoencoders to approximate the posterior distribution, then we can generate high-resolution images from the learnt distribution, paving a way for analysis via synthesis.
Prof. Qingzheng Xu
Speech Title: "Opposition-Based Learning and its Application in Evolutionary Computing"
An opposition concept is both familiar and mysterious at the same time to ordinary mortals like us. However, due to the lack of an accepted mathematical or computational model, until recently it has not been explicitly investigated to any great length in the fields outside of philosophy and logic. The basic concept of Opposition-Based Learning (OBL) was originally introduced by Tizhoosh in 2005. In a very short period of time, it has been utilized in different evolutionary computing areas. This speech covers basic concepts, theoretical foundation, combinations with intelligent algorithms, and typical application fields. A number of challenges that can be undertaken to help move the field forward are also discussed.