Title: Semi-Supervised Learning with Ladder Networks
Abstract: We have recently developed a new type of deep learning architecture called the Ladder Network which has achieved state-of-the-art results in several benchmark tests (http://arxiv.org/abs/1507.02672). Most remarkable improvements over previous results have come in semi-supervised settings which combine state-of-the-art supervised deep learning techniques with unsupervised learning. In this talk I'm going to explain how and why the Ladder Network work. The key to the success has been to device a learning rule which is able to directly learn the inference process needed for hierarchical latent variable models without ever having to write down the probabilistic model. In other words, inference is not derived from a probabilistic model. Instead, it is learned directly from the data.
Bio: Harri Valpola is the CEO and founder of the Curious AI Company (http://thecuriousaicompany.com/) which was founded in September 2015. The company is developing future technologies for AI and is currently doing research on semi-supervised learning, segmentation and hierarchical reinforcement learning. The long-term goal of the company is to develop artificial general intelligence. Dr. Valpola started his academic career as a research assistant of Teuvo Kohonen in 1993 and has been working on machine learning, computational neuroscience and robotics over the years. In 2007 he founded ZenRobotics Ltd. which applies machine learning in advanced robotics, currently recycling by robots.