A desire to extend my knowledge and a great aspiration to contribute ideas for the human-level AI (also refer to artificial general intelligence, AGI) motivate me to pursue a PhD. in Computer Vision. Computer Vision is one of the key disciplines for AI because it helps the computer perceive an object and eventually think humanly, act humanly, think rationally, and act rationally. I am interested in any Computer Vision and its related research topics, such as Computer Vision tasks (Image Classification, Instance Segmentation, Semantic Segmentation, Object Detection, and Object Tracking), emerging Computer Vision data (Point Cloud and Synthetic Data), Computer Vision applications (Automated Driving and Product Recognition), and uncertainty in Computer Vision (Generative Adversarial Network and Neural Tangent Kernels).
Some significant advancements in AI around 2011 sparked my interest. After I enrolled in a master’s program at UCLA, I had an opportunity to broaden my scope and depth of my knowledge, especially in Machine Learning and Deep Learning. After I graduated, I was lucky to join Clobotics Global, which aims to paint an intelligent world through Computer Vision. I was excited to use what I learn from before and implement those ideas into AI-related products. At the same time, I was lucky that many excellent industry leaders could mentor me. During my two years of working as a Computer Vision Scientist, although I am proud of what I have done, I am supposed to get more insights into what I have been doing. Why are insights so important for AI-related work in both industry and academics? Many people claim that Deep Learning is like a black box and believe only know how to play Deep Learning tricks can produce an excellent model. However, as the real AI model steps into people’s daily lives, we need to know those tricks and understand the mathematics behind those tricks during algorithm implementation. For example, our company uses a two-stage model to build the product recognition cloud engine as a Computer Vision Scientist. Some of my colleagues do not understand why we use Faster-RCNN to train the object detection model and use Inception V4 to train the image classification model. In other words, without profoundly understanding many domain knowledge, it is hard to develop the product recognition model with a good performance by selecting numerous feasible algorithms. As a Computer Vision Scientist, I implemented great scientists’ ideas but could not develop ideas as they proposed. Therefore, I decided to resign from the company and considered starting a Ph. D. study to help me become a great scientist. In the next five years of the research study in Computer Vision, I hope to contribute to ideas to impact human society.
WIP…