Research (Selected Topics)
Causal discovery is the problem of learning the causal relationships from observational data. It is a fundamental task in various disciplines of science and engineering. We are working
on computational methods to discover causal relationships from real-world complex data, with theoretical guarantees.
Transfer learning is a research problem that focuses on utilizing previously gained knowledge to solve new problems. For example, when trying to recognize an object at night, we can transfer the knowledge gained when learning to recognize the object during the day. We are working on exploiting causal generative
process of the data to develop more sophisticated machine learning models that can adapt to new domains. In the context of image translation, we are interested in unsupervised learning algorithms to translate images into different domains.
3D Computer Vision
3D computer vision aims to model and understand the visual world by using 3D information. We are interested in deep learning methods for depth estimation from 2D images, 3D reconstruction, and 3D model retrieval. Because the current deep learning-based methods can hardly generalize to new environment, one of our research goal is to develop new deep models that can generalize across different scenarios.