Lifelong Learning for Answerability Education – We propose a deep learning for intelligent machine interaction that aims to be both adaptive and non-adaptive to the interaction of individuals. Based on this approach, multiple machines collaborate to solve a long-running challenge by predicting and learning from input images of interacting individuals, which can be used for the first task of collaborative navigation. Our experimental results demonstrate that as a result of our approach, the two machines converge to a global agreement on the task and that the two machines perform well under different policies.
We propose a deep learning model for modeling the complex semantic object classification problem in robotics. The proposed model is designed to solve the semantic object classification problem: detecting objects from a gallery. Experiments are conducted over a wide range of datasets including both real world and virtual. We demonstrate that the proposed system can effectively predict objects with complicated descriptions. The proposed system can learn the complete object categories of a gallery image from a single image and then fine-tun their semantic labels. Our model outperforms state-of-the-art object classification methods on a variety of robot tasks.
The Evolution of Lexical Variation: Does Language Matter?
Robustness and Generalization in 3D Segmentation of 3D Point Clouds
Lifelong Learning for Answerability Education
Auxiliary Singular Value Classes
We propose a deep learning model for modeling the complex semantic object classification problem in robotics. The proposed model is designed to solve the semantic object classification problem: detecting objects from a gallery. Experiments are conducted over a wide range of datasets including both real world and virtual. We demonstrate that the proposed system can effectively predict objects with complicated descriptions. The proposed system can learn the complete object categories of a gallery image from a single image and then fine-tun their semantic labels. Our model outperforms state-of-the-art object classification methods on a variety of robot tasks.