Page 76 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Computer Science & Engineering

Supervisor: KIM Sung Hun / CSE
Student: PARK Chun Ho / MATH-CS Course: UROP1100, Fall

Neural machine translation is regarded as a promising method of machine translation that could replace traditional
statistical machine translation models. It usually utilizes RNNs because of the sequential nature of processed data, and
the encoder-decoder architecture with LSTM cells has led to good results. This report takes a look at the attention
mechanism that is added on top of the encoder-decoder structure by analyzing the paper “Neural Machine Translation”
By Jointly Learning To Align And Translate. It demonstrates the mathematical methods used to align the output word to
the input sentence, giving attention to parts of input that are most relevant to the current output word, and describes
the results of an experiment that proves the effectiveness of this model.

Supervisor: KIM Sung Hun / CSE
Student: XIA Zihan / COMP Course: UROP1100, Fall

This is an analysis and comprehension about DEEP CODER: LEARNING TO WRITE PROGRAMS, which is a paper written by
Alexander L. Gaunt, Marc Brockschmidt, Sebastian Nowozin and Daniel Tarlow. Their work is about solving simple
competition problems by analyzing input-output examples using deep learning techniques. And my task is to implement
this work using codes, and is still in process. The first thing to do is to find all functions defined in the DSL or write them
by myself, and the second thing to do is to finish the model building by using Tensorflow via Python.

Machine Learning and Chatbot
Supervisor: KWOK James Tin Yau / CSE
Student: FAN Ziqian / COMP Course: UROP1100, Spring
Yang Hsuan Ru / CS UROP1100, Spring

With time pass by, the world is coming closer and closer to the best of machine learning and natural language processing
technology. From dictation input method to intelligent helper, those technologies have come a long way and help us in
several aspects of our life. The purpose of this research was to have a basic understand of such technologies and
develop a chatbot that can help people with their work or daily life. A price searcher bot and a chit-chatting bot are
developed in this project. Those price searcher bot allows users to ask price of stock or digital currency and exchange
rate, and the chit-chatting bot receives user input with speech recognition techniques and generates responses with
natural language generation algorithms.

Deep Learning and Fintech

Supervisor: KWOK James Tin Yau / CSE
Student: KIM Zi Won / COMP Course: UROP1100, Spring
MA, Qiurui / COGBM UROP1100, Spring
SU, Heng / COMP UROP1100, Spring
YANG Yifeng / MATH-SFM UROP1100, Spring

In recent years, deep learning has proven itself as a powerful technique for various applications. These applications
range from object classification from images to forecasting events to natural language processing. In this research
project, we try to apply machine learning and deep learning techniques for predicting stock prices from historical data
only. Specifically, we utilize deep Q networks, a form of reinforcement learning, long short-term memory recurrent
neural networks, a combined approach with both LSTM and RNNs, and convolutional neural networks to tackle the

66