1. ERYA YU
(626) 365-7062 Pasadena, CA 91106 eyyu@caltech.edu
OBJECTIVE
Fulltime Software Development Engineer
EDUCATION
California Institute of Technology, Pasadena, CA, U.S.A Expected Dec. 2016
Master of Science, Electrical Engineering GPA: 4.00/4.00
Core Courses: Relational Database, Machine Learning Data Mining, Communication Network, Networks: Structure Economics,
Functional Programming, GPU Programming
Beijing University of Posts and Telecommunications (BUPT), Beijing, China Sept. 2011 - June 2015
Bachelor of Engineering, Telecommunications Engineering with Management . .. .... GPA: 3.82/4.00 Rank: 2/382
PROFESSIONAL SKILLS
l Language: Java (>5k lines), Python, C/C++, SQL, CUDA, Haskell and UML
PROJECT EXPERIENCE
Machine Learning: Automatic Poem Generation based on HMM Feb. 2016
l Implemented the Expectation-Maximization algorithm for unsupervised learning of Hidden Markov Model (HMM) by
using Python.
l Preprocessed/tokenized the dataset, learned the rhyme, meter and syllables on the entire corpus of Shakespeare’s sonnets
and generated sound poem with the training model.
Machine Learning: Sentiment Analysis Competition in Kaggle Jan. 2016
l Solved the sentiment prediction problem by analyzing a bag-of-words representation of a speech. Adopted feature
selection, TF-idf term weighting and principle components analysis to pre-process the raw training data.
l Improved the prediction accuracy by adopting cross validation to obtain most suitable parameters and conducting
ensemble selections. Beat more than 50% teams in the final competition.
Implemented PageRank Algorithm By MapReduce Framework Feb. 2016
l Found 20 nodes with highest page rank in a web graph containing 500,000 nodes. Enhanced computation speed by
utilizing Amazon Elastic MapReduce framework.
l Further accelerated the method by using heapq and high changing rate at early iterations, reduced around 20% of the total
time spent comparing to the baseline.
GPU Programming: Non-negative Matrix Factorization (NMF) . May. 2016
l Implemented CPU & GPU versions of NMF using C++ and CUDA, realized two multiplicative algorithms.
l Optimized parallel reduction in CUDA by using shared memory, sequential addressing and unrolling loop. The time used
in GPU version is 3x faster than CPU version in Euclidean distance update rule and 6x faster in divergence update rule.
WORK EXPERIENCE
Huixue International Cultural Communication Co. Ltd, Beijing, China .... .... . May 2015 - Aug. 2015
Summer Intern, Product Manager Assistant
l Designed a Customer Relationship Management (CRM) system using Axure for online courses selling.
l Communicated and coordinated across multiple teams including sales, marketing and technology departments. Monitored
development process, tested the CRM system and finished the project on time within two months.
Queen Mary University of London, U.K. ........... Jan 2014 - May 2015
Research Assistant
l Proposed a method using multicoset sampling in frequency domain directly to solve the challenge of high sampling rate
and computation amount in wideband spectrum sensing, saving 21 times 480000-point FFT and 47.5% A/D converters.
l Compared the effectiveness in power estimation when different window functions used, and studied the introduced error.