Yifan Zhao, Ph. D.

Experience

• 

  North Carolina State University

  Aug 2018 - Jan 2023

  Research Assistant

  Project: Image Inpainting with Deep Learning Models Applied for Multi-sensor Satellite Imagery Cloud Imputation -Developed a novel machine learning algorithm for image inpainting with ResNet and multi-image contextual attention. -Formulated a multi-resolution benchmark dataset for evaluating deep learning models for satellite image inpainting. -Developed a computer vision deep learning algorithm with a novel patch-based style transfer module for the multi-resolution image inpainting problem and improved the performance by at least 40% compared to the state-of-the-art methods.Project: Cloud Imputation Deep Learning Model Integration-Developed an end-to-end pipeline for reconstructing millions of images with pre-trained deep learning models.-Extended the pipeline for imputation regarding various types of missing data such as terrain occlusions, shadows, and snow.-Containerized the pipeline as a microservice and deployed it onto the sponsor’s server.Project: Scalable System for Searching Large-scale Multi-Sensor Remote Sensing Image CollectionsConstructed a scalable 3-d R-tree based spatiotemporal profiling system for multi-sensor STAC collections metadata. - Developed a multi-filter content-based neighbor querying system for filtering image attributes of interest such as cloud coverage, spectral, spatial, and temporal overlaps. Project: User Preference as Prior Knowledge in Predictive Models with Regularization Proposed a novel regularization for incorporating ranked user preference as prior knowledge into machine learning methods.- Contributed to Sci-kit Learn with a generic framework for plugging user preference into machine learning methods. Show less

• 

  Convexio Inc

  Jun 2020 - Aug 2020

  Data Science Intern

  Analyze a social science data set of 400K-instance size and provide insight for the supervisor and other marketing staff. Build a dockerized REST-ful Flask application providing potential target recommendations and query service for the social science domain clients and supporting domain expert’s decisions.

• 

  LexisNexis

  Jan 2023 - now

  Senior Data Scientist

  Research data science problems and develop machine learning and generative AI features on various LexisNexis products.-Develop Lexis+ AI, the next-generation LexisNexis primary product. Particularly, develop the RAG-based drafting features, including litigation passages, multi-turn conversation, full litigation documents, upload&draft, ecosystem integration, intelligent handoff, and analysis.-Led the efforts to develop the genAI features, Analyze, for justifying LLM analysis and modification suggestions and presenting options of various alternations and references from public and user’s in-house sources.-Led the efforts to develop features for the interaction between genAI-based products and existing data science products. Experiment with the values of multiple data formulations in such interactions and the handoff options from genAI products to other products.-Develop the finetuning workflow for open-source LLMs on AWS SageMaker with distributed training frameworks.-Improve a concept extraction service for recognizing high-value legal concepts in LexisNexis content and user documents with semantic similarity, LLM, BERT-based classification, etc. Show less