Who Says You Can Trust Deep Learning Now?
Despite widespread adoption, end-to-end gradient-based deep learning models remain mostly black-boxes. On the other hand, human cognition is believed to be able to integrate the connectionist and symbolic paradigms. With a symbolic component, common sense priors can be built into the system. This can help reduce the learning workload and facilitating common sense reasoning, thus leading to a more transparent and trustable learning system.
Ph.D. candidate in Deep Learning at National University of Singapore. Semi-wild machine learning researcher with his human-friendly big AI dream. He is expected to graduate in May 2017 from National University of Singapore (NUS), and was fortunately under the supervision of Tat-Seng Chua and Huan Xu, also closely working with Jiashi Feng and Kian Hsiang Low. He received a Research Master's degree in Computer Science (First Class Honours) with scholarship from University of Otago, New Zealand, luckily advised by Brendan McCane and Lubica Benuskova. He is interested in machine learning. More specifically, his research is focused on deep learning, probabilistic reasoning, reinforcement learning and neural abstract machines. He is especially excited about reducing the gap between theoretical and practical algorithms in a principled and efficient manner.