Eng Hock Lee Email and Phone Number

• Facilitated interdisciplinary technical collaboration between machine learning and materials science experts from 4 research groups on a multi-year project funded by the National Science Foundation (NSF).• Investigated the analytical reasoning capabilities of Large Language Models (LLM) through prompt engineering and introduced strategies for integrating LLM tools into mathematics and machine learning education.• Explored a novel machine learning paradigm called Learning Using Privileged Information (LUPI) for materials discovery applications and demonstrated an average improvement of 27% using LUPI compared to standard supervised learning methods.• Developed a hybrid unsupervised-supervised learning framework to predict material’s properties from a large, unstructured materials database.• Achieved >90% accuracy in identifying van-der Waals heterostructure materials suitable for tunnel field-effect transistors using machine learning.• Investigated the double descent phenomenon in deep learning and provided a theoretical explanation based on Statistical Learning Theory.• Enhanced a deep-learning model for 3D hand pose estimation from 2D images by introducing biological constraint and eliminating occlusion. • Provided assistance to over 100 students in physics and machine learning courses as a teaching assistant.

• Improved cancer cells detection from holographic images of blood samples by optimizing and incorporating cell morphology and optical physics principles into the deep learning model, resulting in a 10-fold reduction in the false positive rate.• Designed a deep learning model for autofocusing holographic cells images to enhance image quality by developing a novel transfer learning strategy using a synthetic dataset for the initial training, followed by fine-tuning with a set of self-labeled images.• Deployed the above models into production, resulting in a significant boost in performance, stability, and image quality.• Implemented model interpretation methods for identifying the most important cell features that affect the deep learning model.

• Secured trade secrets and a patent for developing optimization methods that significantly enhanced the reflectivity of EUV multilayer mirror, with a projected doubling of EUV light intensity output after 10 mirror passes.