Home
About me
Courseworks
Courses

Yiran (Erin) Su

alt text 

Yiran (Erin) Su
Master Student
Electrical and Computer Engineering
University of Texas at Austin
Contact: Email

Education

  • M.S., Department of Electrical and Computer Engineering, University of Texas at Austin, Austin, TX USA, Aug. 2019 - Present

  • B.E. (with GPA 3.8/4.0), School of Data and Computer Science , Sun Yat-sen University, Guangzhou, Guangdong, China, Aug. 2015 - June 2019

Research Interests

  • Natural Language Processing

    • Nerual machine translationg

    • Dialogue System

    • Word representation

    • Deep Learning in NLP

Research Experience

  • Coursework Project for LIN 393: COMP LING: MINIMAL SUPERVSN
    Oct 2019 - Present
    University of Texas at Austin
    Consistency Regularization in NLP

    • Applied consistency regularization methods in natural language processing tasks in order to improve the robustness of the original model.

    • Imported paraphrase perturbation on input text so as to provide a more human-understandable way of perturbation for consistency regularization.

    • Exploring different kinds of text perturbation method and testing their performance of consistency regularization.

    • Exploring different NLP models(CNN, Seq2Seq, etc) and testing their sensitivities towards consistency regularization.

  • Laboratory Research Assistant
    Mar 2019 - Jun 2019
    Sun Yat-sen University, Prof. Xiaojun Quan's research team
    Generalized attention mechanisms research: take nerual machine translation (NMT) task for an example

    • Reconstructed the attention mechanisms on the Harvard NMT framework Open-NMT in PyTorch.

    • Improved the classic attention method on Open-NMT by 0.25 point of BLEU score on wmt-14 en-de dataset. Our paper is still in process.

    • Sloved the problem of small batch size (which often means insufficient computational resources) may effect model performance in NMT tasks by extending steps and warm-up steps, as well as reducing learning rate.

    • Explored different word encoding methods such as Glove, BPE, share-dict BPE and none embedding. Found out that share-dict BPE encoding is the most efficient method when applied to attention-based NMT task.

Work Experience

  • Dialogue System Internship
    Sept 2018 - Mar 2019
    Tencent Inc., Perceptual Intelligence Group

    • Developed a pattern-based framework for a task-oriented Arena of Valor chatbot in weather and music domain in Python.

    • Designed a user log analyzer for the Arena of Valor chatbot in Python, in order to analyze customer stickiness under different usage scenarios in Python.

Project Experience

  • Share My Review
    Sept 2019 - Present
    Cousework project for EE 382V: ADVANCED PROGRAMMING TOOLS at University of Texas at Austin

    • Designed the app as a book review sharing platform, which supports posting a new review of a book category with a photo, viewing reviews under different book categories, finding nearby posts and readers, sharing posts to social media and so on.

    • Applied Flask as back end, HTML + CSS + Bootstrap as frontend.

    • Developed the corresponding Android app with Kotlin.

    • Developed the corresponding React Native app in JavaScript.

  • International Aerial Robotics Competition
    Sept 2017 - Aug 2018
    Computer Vision Group Member

    • Designed an object detection and location system for an aerial robot.

    • Constructed an SVM ground robot detector in C++ for our system by writing a self-implemented Histogram of Oriented Gradient descriptor (HOG descriptor) and applying OpenCV’s related packages.

  • Sun Yat-sen University information platform
    Mar 2018 - Jun 2018
    Leader of the Android Group, Member of the Web Group

    • Designed a framework of the android project as an information publishing platform for all organizations in Sun Yat-sen University.

    • Analyzed the on-campus students’ demand of lacking for a unified information source.

    • Constructed pages for the Android project in Java using Retrofit2 and RxJava, as well as Android Studio SDK tools.

    • Organized Android group meeting per week to control project and implement system improvement.

    • Designed a front-end logic for the web project using HTML5, CSS and JavaScript.

  • The Pacman Game
    2017 - 2018 Academic Year 1st Term
    Coursework Project for Principles of Aritificial Intelligence

    • Implemented the Pacman game based on UC Berkeley’s CS188 course materials.

    • Developed a multi-agent searching strategy using min-max algorithm and alpha-beta pruning in Python.

    • Implemented a value iteration Pacman agent and a Q-learning Pacman agent in Python.

  • Shortest Path for metro transfer
    2016 - 2017 Academic Year 1st Term
    Coursework Project for Data Structure and Algorithms

    • Designed a system to find the minimum number of transfers for a metro network in Guangzhou, China.

    • Implemented Dijkstra Algorithm using C