Emora: Intelligent Companion Chatbot

 Alexa Prize@Amazon: First Place Winner with $500,000 cash reward 

​

Emory Natural Language Processing Research Lab    

​

The research was conducted for “Alexa prize socialbot grand challenge 3”. Our team won the competition by developing a chatbot that achieved an average user rating of 3.8 on a 5-point scale compared to 3.17 achieved by the 2nd, and led an average conversation duration of 7 minutes and 37 seconds; ​​

• Design and construct a database on an AWS Dynamo DB for querying sports-related information; 

• Build a topic handler in Python that create open-domain sports-related conversations using a state machine framework. 

• Implement and improve a question answering deep learning model(BERT) for topic-specific questions over Wikipedia passages, with known question intents and within an ongoing conversation context 

 Unlimited Trace Tutor: Learning Code Tracing With Automatically Generated Programs 

Principal Investigator: Professor Davide Fossati

​

Paper accepted by SIGCSE, First Author 

​

• Develop a back-tracking algorithm to randomly generate blocks of complex Java code and create software in JavaFX based on the algorithm; first iteration of the automatic generation of tracing problems.

• Conduct research trials with college students, compared the novice programmers’ quiz performances before and after they worked with our system, and found an average learning gain of 18%; 

Analysis of a State Machine-based Interactive Dialogue Management System

Thesis Paper that received the highest honor in Computer Science at Emory University

Principal Investigator: Professor Jinho Choi 

​

Abstract: the sports topic handler has been one of the key components in ‘Emora’ an open-domain chatbot system that competes for the Alexa Prize 2020 which is a university challenge for creating the best social bots. This thesis first gives a comprehensive description of Emora’s sports topic handler including its architecture and innovations. These innovations involve effective approaches to opinion-based state transition dialogue management that uses a daily updated database as well as the derivation of engaging conversations on flashing events in sports. Given this topic handler, Emora is capable of making multi-turn dialogues on any game player and team upon request by inferring statistical facts from the database and sharing its own opinions about the latest topics. These unique features help the sports topic handler to become the highest-rated components in Emora in February and one of the highest-rated components of all time........

​

Link to the paper available at the end of the page

Image Segmentation of Ultrasonic Images of Left Ventricles 

​UCSD XuGroup

​

​

• Implement a UNET based Deep learning model with Pytorch and Keras for segmenting Ultrasonic Medical Images of Left Ventricles. Improve the model by adding attention gates to the network.

• Design a data imputation model to impute ventricle volumes of images that has inferior quality.

GoHousin

• Lead a group of visual designers and programmers to develop a student-affordable apartment searching the website, utilizing agile project management; 

• Construct the website’s front-end interfaces with HTML, CSS, JavaScript; 

• Implement an embeddable forum to build an online community for users.