Yujia Liu is a graduate of Georgia Institute of Technology with a Master's degree in Electrical & Computer Engineering and a minor in Computer Science. He has work experience at Sogou, Inc. where he modified an RNN model to GRU and ported it to GPUs. Some recent projects include a personalized recipe recommendation system using ingredient data and clustering, displaying the Mandelbrot set using GPU computing, and implementing image FFT across CPUs using MPI and Pthreads. His skills include C/C++, Python, Matlab, frameworks like Cuda and MPI, and background in machine learning and systems & control.
1. Yujia Liu
Resume
639 Geary St Apt 1408
San Francisco, California 94102
(404) 662 7726
liuyujia.victor@gmail.com
Education Background
2014–2016 Master in Electrical & Computer Engineering, Georgia Institute of Technology , 4.00/4.00,
Minor in Computer Science.
Selected Courses: Machine Learning, Data Visual & Analytics, Web Search & Text Mining, Advanced
Programming Techniques, Computer Vision, Statistical Techniques for Robotics.
2010–2014 Bachelor in Control Science & Engineering, Zhejiang University, China, 3.81/4.00.
Work Experience
May.–Jul.,
2016
Word Prediction with GRU/LSTM Model, Sogou,Inc., C++/Cuda/MPI.
Modified an existing RNN model to GRU version to keep more information in long sequences.
Ported algorithm on GPU, used 8 GPUs on 4 machines to greatly increase the overall throughput.
Achieved target word probability of 0.20 when the vocabulary size is 8700.
Recent Projects
Sep.–Nov.,
2015
Smart Meal Planner - A Personalized Recipe Recommendation System, Python.
Requested and cleaned the ingredient data of more than 30000 recipes retrieved from Yummly API.
Built a recommendation system with feature vectors defined by ingredients and tastes, clustered the
data to greatly reduce computation of finding the nearest neighbor.
Nov., 2015 Mandelbrot Set Display with GPU Computing, C++/Cuda/OpenGL.
Set up CUDA environment to compute whether a point on the complex plane is belonged to the
Mandelbrot Set and achieved linear speedup.
Used OpenGL to visualize the Mandelbrot Set and implemented the zooming in and out function.
Oct., 2015 Image FFT in Multiprocessing & Multithreading Programming, C++/MPI/Pthreads.
Assigned 16 CPUs/Threads to compute row and column FFT separately.
Used MPI/Pthreads to send and receive partial information between the 16 processes/threads.
Implemented Danielson Lanczos approach to reduce the running time complexity from N2
to NlogN.
Jan.–May.,
2015
Optimization Methods in Training Neural Networks, C++/Eigen.
Built a three-layer stacked denoising auto-encoder to classify handwritten digits in MINST and CIFAR.
Implemented three variants of gradient descent method and compared their performance in terms of
accuracy, convergence rate and running time.
Nov., 2015 Face Detection with Sliding Window, Matlab.
Extracted HoG feature from the training images from Caltech Web Faces and SUN scene database.
Applied support vector machine with linear and Gaussian kernel on CMU+MIT dataset and achieved an
accuracy of 0.924 using the multiscale sliding window approach.
Skills
Coding C/C++, Python, Matlab, SQL
Frameworks Cuda, MPI, Pthreads, D3.js, Node.js, OpenGL, OpenCV
Background Machine Learning, System & Control