I am interested in researches related to Computer Vision, Graph Data Extraction using machine learning Techniques.
I am interested in researches related to Video/Image processing, streaming and management.
I am interested in researches related to robot localization and navigation algorithms.
Visual Common Sense is fully utilized by human and even other creatures all the time. Some of the skills of acquiring environment knowledge were so common that might have been overlooked.
We believe that the ability to make intelligent assessments of where things might be, even though they are not visible, is a critical and ubiquitous capability shared by humans and other intelligent creatures.
In this project, we aimed to explore the concept of visual common sense, and in particular, help to define what it means to perform successful inference about what we cannot see, conditioned on what we can see.
Consider you have a library of aging old videos taking a significant amount of space of the data server, and you want to cut the size of the files down and you don't really have time and motive to go through and process all of the videos. But you really need more space for newer data. In this case, we will need to cut the size of the video files to free space for new data.
A smart and efficient batch algorithm that can be easily applied to large amount of data with minimum amount of human intervention is crucial for such task.
We believe that such attempt on development of algorithm for processing large amount of visual data will have a high demand during this time of information explosion, building upon the combination of successful computer vision capabilities such as object detection and image segmentation, and Natural Language Processing techniques such as word representation and image captioning.
People have been trying to train machines to understand human natural languages for several decades already. We have been changing the ways how we represents words, we feed novels and essays to them and hope that some day, the machines can understand will understand what we are saying. However, the programming languages, invented and written by human, was developed to be read and understood by machines. Compared to human natural languages, programs are more logically structured with much less vocabularies. We believe this can be a good starting point for us to try to dig into research in natural language processing.
Currently auto-completion for scripting language are mostly based on naive conditional and search algorithms. A good auto-completion algorithm that is independent to compilers will be extremely helpful to boost productivity and to step forward in Natural Language Processing researches.
In this project, we are trying to develop a light-weight automated lecture capturing scheduling and managing system that supports for LMS platforms with LTI Specifications and investigate how multimedia content delivery systems support and enhance teaching and learning at a distance and in the classroom. We are also evaluating the application of multimedia technologies to collaborative work and outreach to learning and other communities.
Robotics is the science of perceiving and manipulating the physical world through computer-controlled devices. Robotics systems are situated in the physical world, perceive information on their environments through sensors, and manipulate through physical forces.
Intelligent autonomous mobile robots need to know where they are, in order to function correctly. For example, a self-driving car needs to know which street, and which intersection it it at, in order to plan when and which turn it should take next. While some robots have access to GPS readings, such GPS readings are still not sufficiently accurate to make accurate decisions - the error of a commercial GPS is greater than the width of a driving lane, and the decisions taken by a self-driving car need to account specifically for not only which lane it is in, but exactly where in the lane it is.
In this project, we silulated the sources of uncertainty for the problem of mobile robot localization, and one popular algorithm for mobile robot localization, called Monte-Carlo Localization, works on robots.
Built a 2-Dimensional puzzle game with real-time multiplayer collaboration functionality. Developed using Qt Framework with applications of Singleton, Observer and other design pattern. Delivered fully functional game client and server
Worked on a mobile application that records time arrangement of the user with statistical analysis and evaluation over time distributions on certain behaviors. Implemented the whole application including user interface, Service layer and Data management. As of Feb 2017, the project has about 30 Alpha testers from various countries
Built an outsourcing hiring information management system with chatting functionality via a chat bot using Google Framework that based on QQ, a social media platform by Tencent. The application resolves the communication barriers between employers and employees. Pushed to production and have about 4000 clients and maximum of 200 active users within 24 hrs
Worked on a simulation program to simulate, benchmark and visualize Probabilistic Robotics models. Analyzed performances of Monte Carlo Localization and Particle filter for robot localization
Implemented modules that fetches images from online databases, handles image data preprocessing, image pattern classification and performs data extraction. Classifies XRPD Graphs and Raman Shift Graphs and extract peak values for estimating Crystal form based on both self-generated and obtained datasets with size of over 1,000,000. The trained CNN based graph recognition model achieved an accuracy of about 90 percent and the data extraction utility achieved an error rate of ±0.01 cm given a letter-size graph
Implementation of an optimal solution of portable automatic-scheduling lecture capturing and processing management system that supports for LMS platforms with LTI Specifications. Responsible for the capturing scheduling and management system, and back-end modules that communicate with the presentation front-end and the data storage server. Expecting a product that handles scheduling, capturing, processing and streaming lecture and other visual-aid sources that runs on a portable, inexpensive platform Raspberry Pi
Research of proposing benchmarks and experimental algorithms analyzing co-occurrences on evaluating the probability of multiple target images being in the same scene as the reference image (Half&Half Benchmark). Optimized algorithms that classifies images into Topic-model-like distribution using natural language processing techniques from features extracted from deep learning network layers. Intended applications include feature-based text recognition algorithm, unsupervised learning over classification model, and also vision analysis on robotics localization and navigation
Checkout the link above for my music studio!