FST Graduate Research Topic

Research Topic: Statistical and Deep Learning for Text and Image Analysis
Suggested By: Dunwei (Grant) Wen
Last Update: September 9th, 2014

This research project focuses on natural language text and image analysis based on statistical and deep learning techniques, for recognizing semantics or meaningful relations from unstructured big data, for better understanding of the data and serving the users. This project aims to develop a mixed natural language text and image analysis system for an online social network environment, where plenty of articles and comments including both text and images are increasingly produced by its users. Several predefined events and semantic relations useful for the stakeholders will be extracted, identified and recognized against the mixed data by using statistical and deep learning algorithms, and then visualized in an interface as real time report. The key research problems include how to choose suitable language and image processing models, how to design effective statistical and deep learning algorithms and programs for analyzing the unstructured data, and how to make use of higher level semantic information in the analysis. Moreover, some additional information associated with multiple modals, will ideally be jointly modeled in the learning process and further leveraged for analysis. The prototype system of the project should be able to perform effective text and image analysis with expected outputs for the purpose of proof of concept. The required skills/background: * Strong background in artificial intelligence, including machine learning, natural language processing and computer vision. * Strong interest in unstructured big data analysis. * Excellent skills in developing Java and C++ programs with advanced data structures and algorithms. * Familiar with Linux programming and development environment. Planned research tasks: * apply open source NLP and image processing tools such as OpenNLP, StanfordNLP and OpenCV for text and image processing and feature extraction. * design selected statistical and deep machine learning model and algorithms for analyzing multimodal text and image data according to the defined analysis tasks * perform experiments for evaluating the designed algorithms and tools. * develop a prototype system for demonstrating the integrated tasks of the application. * writing progress and development reports, and research papers. This research topic is suitable for Project or Thesis.