Bingjia Wang
EMAIL: adjbjz@r.postjobfree.com PHONE: +1-347-***-****
LinkedIn: https://www.linkedin.com/in/bingjia-wang-a6271b139/ GitHub: https://github.com/wbj218 Local Address: 808 Dryden Rd, Apt B, Ithaca, NY 14850 Permanent Address: No.5 Jianguomen North Street, Room 1511, Beijing, China 100005 EDUCATION
Cornell University – Ithaca, NY Dec 2020
Master of Engineering in Electrical and Computer Engineering Overall GPA: 3.91/4.00
• Relevant Coursework: Bayesian Estimation and Stochastic Optimization; Data Mining and Machine Learning; Natural Language Processing; Computer Vision; Datacenter Computing; Embedded Operating System; Operating System; Data Structure Rensselaer Polytechnic Institute – Troy, NY May 2019 Bachelor of Science in Electrical Engineering, Minor in Mathematics Overall GPA: 3.96/4.00 HONORS & CERTIFICATES
Dean’s Honor List on All Semesters, RPI
Summa Cum Laude, RPI (May 2019)
RPI Harold N. Trevett Award (May 2019): Awarded annually to a senior for outstanding scholarship in EE. Bose Lean Enterprise Yellow Belt Certificate (June 2018) SKILLS
Python (TensorFlow, Scikit-learn, PyTorch), Java, Node.js, C, R, SQL, AMPL, MATLAB & Simulink, Cadence WORK EXPERIENCE
Journeyman Fellowship Research Intern, U.S. Army CCDC Army Research Laboratory May-Aug 2020 Internalizing Succinct Gaussian Processes by Integrative Policy Search in Reinforcement Learning
• Used compressed online Gaussian Process to estimate the environment transition dynamics and the state occupancy measure.
• Used Gaussian Quadrature to compute the integrative policy gradient w.r.t. the estimated state-action occupancy measure.
• Update the policy by extending the Palyak-Ruppert Averaging.
• Experimented on a few OpenAI environment and observed an order of magnitude reduction on sample complexity during training. Electrical Engineer Co-Op, Bose Corporation - Framingham, MA Jan-Jun 2018
• Used C and Matlab to implement different embedded sensor systems to develop algorithm for headphone on-off head detection.
• Implemented cap-touch system on headphone earcup to detect gestures.
• Developed test procedure, collected and analyzed data for on-off head detection algorithm improvement. RESEARCH & PROJECTS
A Markov Decision Process Approach to Active Meta Learning, Cornell – Ithaca, NY arXiv Link Aug 2019-May 2020
• Built a stochastic scheduler on top of existing deep meta learning algorithm to actively select samples on which to train from different meta-training sets, in order to improve the sample efficiency.
• Developed scheduling schemes based on multi-armed UCB and Gittins indices algorithms to exploit covariates within subset.
• Further exploited the dependencies between different training subsets by modeling the problem as Markov Decision Processes.
• Experimented the schemes on 3 datasets and showed that they could improve training efficiency and maintain high accuracy. NLP Projects, Cornell – Ithaca, NY Aug-Dec 2020
• Question-answering: Developed 2 question-answering models using DrQA and pretrained Bert model (in PyTorch) to answer questions and predict questions’ answerability in SQuAD 2.0 dataset.
• Emotion Classification: Used FFNN, RNN and LSTM (in PyTorch) to classify different emotions using Tweeter posts dataset.
• Sequence-tagging: Built features and used HMM and MEMM to identify propagandistic spans of text for tokens.
• Real/Fake News Classification: Use n-gram language model with smoothing and Naïve Bayes to classify real and fake news. Machine Learning Projects, Cornell – Ithaca, NY Aug 2019-Dec 2020
• Drug MoA Prediction: Used LightGBM and neural network (multi-seeds FFNN and ResNet) to predict MoAs on different molecules.
• Modulation Prediction: Used Python with sklearn and TensorFlow to implement AdaBoost with Random Forest, CNN and ResNet to predict the signal modulation schemes.
• Divorce Prediction: Used R to implement various parametric and non-parametric machine learning methods (LR with regularization, SVM, trees and random forest, PCA with nearest neighbors) to predict divorce; Used LASSO, forward and backward selection, data shrinkage to identify important features and interpret the findings. SLAM Mapping Robot, Cornell – Ithaca, NY PROJECT WEBSITE Aug-Dec 2019 Used Raspberry Pi, piTFT, IMU, wheel encoder and ultrasonic ranger to design and build a SLAM mapping robot to achieve area dimension and altitude measurement functionalities and show real time result on piTFT screen.
• Used IMU with low-pass and complementary filters to compute robot heading direction and altitude measurement.
• Used wheel encoder and ultrasonic ranger to achieve distance measurement and edge detection.
• Designed algorithm to localize robot and show the robot position and altitude measurement on piTFT in real time.