San Francisco Bay Area
Designing and implementing Deep Learning architectures and data processing pipelines for medical imaging applications at ACIST Medical Systems
- Developed multi-threaded data processing and augmentation pipelines for training various Tensorflow-Backend CNN, Regression, and GAN Keras models. - Designed Generative Adversarial Network architecture to generate 3x3 Color Correction Matrices for augmenting image colors. - Built Regression Network that outputs Blood Volume, Oxygen Saturation, Melanin Content, and Skin Refractive Index information for each pixel of a supplied hyperspectral image of human skin. - Constructed novel CNN architecture to negate ambient illumination from 15-channel input image by generating 3-channel image in CIELAB L*a*b* color space using per-pixel embeddings generated by in-house pretrained Regression Network that I developed.
- Used Openface Python library to extract features from medical patients - Constructed a Random Forest model in Python to determine which facial features are most indicative of a patient's potential risk of difficult intubation
- Conducted a self-guided research project to use deep learning to improve tissue labeling done by pathologists during the prognosis of bladder cancer - Produced automated alternative to manual labeling, which takes 24 hours for an individual biopsy - Trained a CNN based on the U-Net paradigm to perform per-pixel multi-class instance labeling on given bladder biopsies - Model accuracy measured to be 90% and can label each pixel of a given input (5-18 GB per input) in 5 minutes, as opposed to 24-hour manual labeling that is the industry norm - Continued as a part-time remote researcher after the official internship period ended in August
- Headed a team of UCSC Capstone Project students to work with Continental’s ADAS team to develop the foundations of a Simultaneous Localization and Mapping (SLAM) system for driverless vehicles - Developed a subscriber and publisher system in ROS for real-time streaming of radar data from test rig - Implemented Extended Kalman Filter in Python to reduce noisy localization ((This project was sponsored by Continental and UCSC through the Senior Design Capstone Project program))
As a grader for this class, I graded assignments written in the following languages: -Scheme -OCaml -Smalltalk -Perl -Prolog