Engineering Data

Cheng He

805-***-**** **** S Grand Ave, Apt ***, Los Angeles, CA 90015

*********@*****.*** ● https://www.linkedin.com/in/chenghe0430/

EDUCATION

Columbia University School of Engineering and Applied Science New York, NY

M.S. in Mechanical Engineering Expected: Dec 2019

UC Santa Barbara School of Engineering Santa Barbara, CA

B.S. in Mechanical Engineering Oct 2014 - Oct 2018

TECHNICAL SKILLS

Machine Learning Models: Logistic Regression, DNN, Deep Q-Learning

Programming Language: Python, ROS, MATLAB, MySQL, SAS, R

Relevant Courses: Artificial Intelligence, Data Science, Control System Design, Computer Aided Manufacturing

Robotics: Motion Planning, Cartesian Control, Forward Kinematics, Numerical Inverse Kinematics, Rapidly-Exploring Random Tree (RRT), Extended Kalman Filter

PROJECT EXPERIENCE

Othello Programming, Artificial Intelligence Sep 2019 - Dec 2019

Collected possible actions from the Othello board provided and called one player as min player, the other as max player

Chose the move each player would take to maximize their scores with utility function

Revised the code with minimax algorithm (alpha-beta pruning) by adding the alpha value and beta value

Added limit level to increase the calculating speed of the program

Observed the results of the program playing with AI agent on a 4x4 board, 6x6 board, and 8x8 board;

Robot Simulation, Robot Learning Jan 2019 - May 2019

Collected training data from ROS service, which took an action and got the next state as response

Trained the DNN with PyTorch, taking the current state with action as input and next state as output

Minimized DNN errors by modifying DNN structure, tuning hyperparameters, and optimizer

Implemented final DNN to simulate robot reaction and compared error (less than 1% error)

Australia Humidity, Data Science Jan 2019 - May 2019

Performed data-cleaning on Australia weather data and selected features based on the correlation to humidity

Separated cleaned data set into training set, validation set and testing set

Trained training set using logistic regression model to predict humidity

Tested on testing set and evaluated the model performance with error between predicted data and real data (2% test error)

Simulated Cartpole Robot, Robot Learning Jan 2019 - May 2019

Established an action network and a target network using PyTorch to implement a Deep Q-learning algorithm

Stored transitions, which includes action (force and direction to push the cart), state (cart displacement, angle of pole with respect to vertical, speed of cart, and angular speed of the pole), next state, and rewards (if pole not fall, reward=1; otherwise, reward=0), from the given ROS service

Sampled random minibatch of transitions from stored transitions and performed Gradient Descent on action network for each episode (each episode has several time steps)

Updated target network to be equal to action network every k episodes to prevent overfitting

Maximized rewards by tuning the hyperparameters of the Deep Q-learning Network, time step of one episode, and k

Evaluated the trained robot by testing if the robot can keep the cart displacement within 1.2, and pole angle within 6 after 200 time steps

Contact this candidate