Jieyuhan Zhu

• Used PyTorch to establish 2 deep-learning models based on U-Net and GAN architectures, resulting in an 8% improvement in testing accuracy and accelerated model training speed. • Preprocessed, normalized, and denoised DICOM data of 400 pediatric patients into 2 formats for modeling, increasing both U-Net and GAN models’ testing accuracies by 15%. • Achieved comparable performance to CT-based AC, proved feasibility of DL alternative, promoting long-term health benefits for pediatric patients. • Co-authored and published paper Deep Learning-Based CT-Free Attenuation and Scatter Correction for Pediatric Whole-Body PET Imaging in The Journal of Nuclear Medicine.

Spatial Propagation Learning and Segmentation for 3D Medical Images 1. Converted the medical images into 2D slices and built a CNN network to extract feature pyramids from adjacent slices. 2. Designed an optical flow propagation neural network, inputting features of adjacent 2D slices to determine the optical flow relationship between adjacent slices. 3. Established a 3D spatial optical flow propagation neural network, implemented obtaining the segmentation of an entire 3D medical image from the optical flow propagation of a single annotated 2D slice, enhanced the efficiency of 3D segmentation. 4. Constructed an affinity matrix neural network to discern pixel correlations within each 2D slice, refining the initial coarse segmentation outcome, increased accuracy (DICE) between the training results and ground truth by 17%, achieving a final average segmentation accuracy (Dice) of 86%.

• Analyzed dating app data using Excel (formulas, VLOOKUP, pivot tables) to assess the most profitable app categories and core user segments, compare different monetization models, and proposinge a strategic roadmap for the company’s dating app sector based on these insights. • Analyzed data on in-app advertising (IAA) apps, including categories, advertising methods, ARPU, and payback periods, to identify most profitable app type, optimize advertisement placement, enhancing the commercialization of IAA apps commercialization. • Created 80+ visualizations such as pie charts, bar charts, and bubble charts using ThinkCell in PowerPoint and compiled a report to deliver business insights, providing data analysis for the company’s commercialization strategy. • Collaborated with teams to coordinate needs, synthesize findings, and deliver insights, ensuring alignment with goals.

• Collected and preprocessed 360 Mini-Robot gripper coordinates and corresponding joint angles using MySQL to identify outliers and missing values, storing cleaned data for modeling. • Performed model selection using Python to build and evaluate multiple machine learning models, identifying a multivariate regression model with the highest accuracy to enable the robot to adjust joint angles for precise gripper positioning. • Design an operating system using Python Tkinter, enabling the MinRobot to automatically operate its joints based on user-defined gripper path input, enhancing production line efficiency by assisting workers with product handling.