Liqun Yang

School of Computing and Information Sciences


Lecture Information:
  • June 3, 2020
  • 4:00 PM
  • Zoom: https://fiu.zoom.us/j/92159710511

Speaker Bio

Liqun Yang is a Ph.D. candidate of Computer Science in the School of Computing and Information Sciences at Florida International University (FIU), under the supervision of Dr. Wei Zeng. In the past 3 years, Liqun focused on the intelligence prediction and pose detection based on the neural network. His research interests are in the deep learning for computer vision and analysis of the neural network.

Description

As the research of neural networks going further and further, the nature of the learning of the neural network attracts more and more attention. Although some neural networks earn great success in the solution for some hard problems, people still cannot answer the question of why these models can do that and what does the model learn in the training process. In this research, we plan to provide a new method to describe the changes in the training process of the neural network based on the optimal transpose theory. Comparing with the performance-based evolution analysis, this method has two advantages: (1) Provide a meaningful differentiable curve under a defined metric; and (2) Reflect the nature of the learning process. Some preliminary experiments have proved the effectiveness of this method. We will futher expand the scope of our experiments to verify the correctness of our theory and imply it to the dataset evaluation and self-supervised neural network implementation.