Research and Intern Experience
Game 2048 AI designing based on supervised learning and CNN
Facilitating game board initialization, player move data storing and training, prediction based on trained AI by constructing respective objects. Construct CNN with 5 convolutional layers and 3 dense layers. Running 300000 games with super AI, and sample 1000 moves from each game for supervised training. Evaluate trained model performance according to 50 games average score, and build website app that display model playing on graph interface.
Research on Camera Dependency in Image Recognition
Apply SDK to control Basler acA1920-40gc industrial camera to sample image collection under different distances, lighting and vehicle speed Build models to adjust gain, black level, gamma coefficient, white balance to improve image recognition Simulate camera ISP effect with SDK models based on street shoot collected data.
Biomarker and Endpoint Assessment to Track Parkinson’s Disease
Collect 3-axis data and analyze in time domain, frequency domain, and statistics. (R) Do modeling, CBDA analytics, forecasting and cross-validate on the results.
