UROP Project

AI-Driven Data Recovery and Reconstruction for Scientific Discovery

Artificial Intelligence; Machine Learning; Model Inversion; Materials Discovery

Research Mentor: zp23e@fsu.edu ZHIXIN Pan, Dr.
Department, College, Affiliation: Electrical and Computer Engineering, FAMU-FSU College of Engineering
Contact Email: zp23e@fsu.edu
Research Assistant Supervisor (if different from mentor): Amberbir Alemayoh Mr.
Research Assistant Supervisor Email: ama25d@fsu.edu
Faculty Collaborators:
Faculty Collaborators Email:

Looking for Research Assistants: Maybe one more
Number of Research Assistants: 1
Relevant Majors: Computer Science, Electrical and Computer Engineering, Data Science, Materials Science, Chemical Engineering, Chemistry, or related STEM majors. Students with strong programming or machine learning interests from other majors are also welcome.
Project Location: CAPS
Research Assistant Transportation Required: No, the project is remote
Remote or In-person: Partially Remote
Approximate Weekly Hours: 5-10, Flexible schedule (Combination of business and outside of business. TBD between student and research mentor.)
Roundtable Times and Zoom Link:
Not participating in the roundtable

Project Description

Modern artificial intelligence models can learn complex patterns from large datasets, but an intriguing question remains: how much information about the original training data is retained inside a trained model? This project will explore this question through model inversion and data reconstruction, a growing area of AI research that aims to recover or approximate information learned during model training.

The student will investigate machine learning techniques for reconstructing data from trained prediction models. We will begin with controlled experiments in which models are trained on known datasets and part of the original training information is hidden. The student will then implement and evaluate model inversion methods to determine what information can be recovered from model outputs, internal representations, or other accessible model information.

The project will examine factors that affect recovery performance, including model architecture, data representation, available model access, and reconstruction algorithms. Recovered samples will be evaluated for accuracy, similarity, diversity, and usefulness in downstream machine learning tasks.

As a scientific application, the developed methods will be evaluated using datasets and prediction models for polymer and materials research. The long-term goal is to understand whether trained scientific AI models can serve as an additional source of recoverable data for AI-enabled scientific discovery.

Research Tasks: The research assistant will:

• Review introductory literature on model inversion, data reconstruction, and related machine learning techniques.

• Learn to train and evaluate baseline machine learning models using Python and modern ML frameworks.

• Reproduce selected model inversion or data reconstruction methods from existing research.

• Design controlled experiments to study how much training information can be recovered from trained AI models.

• Investigate how factors such as model architecture, data representation, and level of model access affect reconstruction performance.

• Evaluate recovered data using reconstruction accuracy, similarity, diversity, and downstream machine learning performance.

• Apply the developed methods to scientific datasets and prediction models, with polymer and materials data serving as an initial application domain.

• Analyze results, prepare figures and tables, document findings, and contribute to a research poster.


Skills that research assistant(s) may need: Required:
• Basic programming experience and willingness to learn Python.
• Interest in artificial intelligence, machine learning, or data science.
• Ability to work independently, document experimental results, and communicate progress regularly.

Recommended but not required:
• Experience with Python, PyTorch, TensorFlow, or other machine learning tools.
• Coursework in machine learning, data science, algorithms, statistics, or related areas.
• Familiarity with neural networks, optimization, or scientific computing.

Prior research experience and prior knowledge of the scientific application domain are not required. Training and guidance will be provided.