Jiayuan Huang

• Focused on uncertainty quantification of subsurface properties in unconventional shale formations to enhance resource assessment and decision-making. • Specialized in leveraging generative AI and Geostatistics to develop geologically consistent subsurface models, bridging the gap between data-driven innovation and traditional simulation methods.

Project Title: Optimizing Text-to-Image Generative AI for Subsurface Modeling • Proposed and implemented innovative CLIP model training strategies, enhancing domain-specific prompt comprehension and improving the quality of generated images for subsurface modeling. • Fine-tuned a Large Language Model (LLM) using a domain-specific text dataset to ensure alignment with geological terminology and contextual accuracy. • Designed and developed an algorithm to assess the inference accuracy of CLIP models for domain-specific prompts, enabling precise evaluation of model performance. • Conducted extensive experimentation and evaluations, achieving substantial improvements in model precision, computational efficiency, and geological consistency.

Project Title: Developing Text-to-Image Generative AI for Subsurface Structure Modeling • Focused on designing the foundational framework for text-to-image generative AI models capable of simulating geological subsurface structures based on domain-specific textual descriptions. • Curated high-quality training datasets with pairs to ensure geologically accurate outputs and robust model performance. • Conducted baseline experiments leveraging CLIP, analyzing model behavior to identify critical areas for improvement and enhance output accuracy and consistency.