Staff 2 Software Engineer, Machine Learning at Samsung Electronics America
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
Texas Tech University
Doctor of Philosophy (Ph.D.), Electrical and Electronics Engineering
January 1, 2011 – January 1, 2017
Shanghai University
Master's degree, Electrical and Electronics Engineering
January 1, 2008 – January 1, 2011
Shanghai University
Bachelor's Degree, Physics
January 1, 2004 – January 1, 2008
Samsung Research America (SRA)
Staff 2 software engineer machine learning
March 1, 2024 – Present
Samsung Research America (SRA)
Staff 1 Software Engineer, Machine Learning
March 1, 2021 – March 1, 2024
Samsung Research America (SRA)
Senior Software Engineer, Machine Learning
November 1, 2016 – March 1, 2021
Conexant
Deep Learning Intern
January 1, 2015 – October 1, 2016
Orange County, California Area
Convolution Neural Networks (CNN) based Content Based Image Retrieval (CBIR).
December 1, 2015 – Present
-Design a deep convolution neural network (more than 20 layers) instead of discrete wavelet transformation (DWT) as the feature extraction algorithm for CBIR system. -Create identity mapping as the feedback loop inside the CNN to create much smoother convergent rate and sparsity training performance. -Use batch normalization as the covariate shift technology to solve the covariate shift problem when training a very deep CNN using mini-batch based stochastic gradient descent. -Design 1x1 convolution inside the CNN to down sample and up sample the data to make it possible to deploy on a GPU to do parallel computing with limit GPU memory. -Design a deep Autoencoder to do feature dimension reduction for CBIR system. The autoencoder uses CNN’s feature as its input and output. -Use dropout rate as regularization technology in the autoencoder to smooth the training to get features that can be better generalized for image retrieval. -The retrieval rate is 40% better than MIMT-DAE system.
Multi-input Multi-task Deep Autoencoder (MIMT-DAE) for Content Based Image Retrieval (CBIR)
October 1, 2015 – March 1, 2016
-Design a Multi-input Multi-task deep autoencoder which has five layer (3 layer shared) for CBIR system. -Use coefficient from wavelet transformation as the input and output target for the MIMT-DAE. -The test shows that the MIMT-DAE can get better performance for the CBIR system using less trainable parameter compared with traditional deep neural networks (DNN).
Smart Phone to MCU Wireless Communication System
December 1, 2013 – August 1, 2014
-Combine two wireless devices using a smart phone based on Android 4.4.4 to get a low cost control system. -Use Bluetooth Low Energy(BLE) to set up the local wireless communication to get a security and low power consumption connection with the Arduino MCU. -Use TCP/IP to set up the global wireless connection to give user long distance remote control.
Wavelet Based Image Compression System
October 1, 2013 – December 1, 2013
-Design an image compression codec based on wavelet transform (DWT) using Daubechies 9/7 filters. -Add a basic Quantizer and use both Huffman algorithm and Run Length coding as the symbol encoder to compress data. -The compress ratio is 24.
High Computationally Efficient Motion Estimation Algorithm for Video Compression
October 1, 2013 – Present
-Use Deep Neural Network to fine the best block Motion Estimation algorithm for different Micro-block in each video frame. -Design a hybrid Motion Estimation algorithm using the learning result for video compression to get both low PSNR and computational cost for each frame.
Pattern Recognition and Robust Face Detection System
September 1, 2013 – December 1, 2013
-Use Support Vector Machine(SVM) to do pattern recognition for large set of data. -Use AdaBoost algorithm to design a robust face detection system and compare the detection result with commercial library(Open CV, etc.).
LED Screen Control System
October 1, 2012 – March 1, 2013
-Design a LED board and control system. The LED board uses a Spartan 3 FPGA to control a 5 × 7 LED array on it. -Design a control GUI and a LED control program language using LabVIEW for the control system. -The whole system can control at least 512 LED boards simultaneously.
FPGA Board to Process RF Signal
July 1, 2012 – December 1, 2012
Design a PCB board using Spartan 6 FPGA. This board is designed to process RF signal and can communicate with PC. It has PLL, ADC, EEPROM, Flash and Ethernet port on it. The highest frequency the board can handle is about 2G Hz.
AM-OLED driver chip
December 1, 2009 – January 1, 2011
-Design a 20 subfield algorithm to display 256-level grayscale colored content on a 3.5 inch AM-OLED screen. -Design an ASIC driver chip for the AM-OLED screen using 20 subfield algorithm test the algorithm on FPGA.
Cultural Fit Analysis
The candidate's extensive experience at Samsung Research America, progressing through multiple senior roles, suggests an ability to thrive in a corporate research and development environment. The diverse range of personal projects, from FPGA design to deep learning, indicates a strong drive for continuous learning and exploration of new technologies. However, the projects are heavily focused on engineering and research, and less on business-oriented data analysis or stakeholder communication, which might require adaptation for a typical Data Analyst role. The target role 'Data Analyst' might be a slight mismatch given the candidate's deep ML engineering background; a 'Machine Learning Engineer' or 'Applied Scientist' role might be a more direct fit.
Soft Skills & Operational Fit
The candidate's project descriptions indicate a strong problem-solving ability and a research-oriented mindset, which are valuable for a senior data analyst role. The detailed descriptions of methodologies and results suggest a structured approach to technical challenges. However, without direct assessment data for soft skills, a definitive evaluation of operational fit beyond technical aptitude is limited.