Study
PLCO
(Learn more about this study)
Project ID
PLCO-1275
Initial CDAS Request Approval
Jul 19, 2023
Title
Exploring the Viability of New Technology to Quickly Determine All Feature Contributions, Feature Value Contributions, Feature Interactions, and Feature Value Interactions that lead to cancer.
Summary
Our company has developed new technology that computes accurate and exact Shapley values for any sized dataset prior to any model generation. The implications for this are paramount, in reducing Exploratory Data Analysis from 16-40 hours to 1 hour while arriving at exact values, opposed to current day descriptive statistics. Demonstrating the exact ground truth to aid in the process of model selection in a much better way than AUC, accuracy, confusion matrices, or permutation importance. This technology is primarily useful and applicable in areas utilizing machine learning and artificial intelligence in high risk arenas such as medical, finance, government, and utilities applications. This is very important for any field where explaining what the models are doing and why is key. We are first starting with Pancreatic Cancer research because our specific collaborator is working in this field. However, this study will form a baseline for all other structured data research projects involving cancer research. Our team is comprised of a Machine Learning/AI Engineer, and a PhD from the University of California Berkeley, in Computer Engineering. Our breakthrough has been the product of 20 years of work and we are eager to collaborate and demonstrate its inner works with those actively working in the Pancreatic Cancer research.
Aims
Our goal is to conduct a study of the viability of this technology on already known practical dataset in the Pancreatic Cancer research field and present our findings to key partners in the medical field for use in their own research. This will progress the state of the art in the field of cancer research by providing a never before seen method to data analysis, model selection, and model/data explainability for deployment into production environments. Demonstrating this capability while providing the insights to our collaborative partner based on this new technology will further advance the area of cancer research and create a paradigm shift for how machine learning is applied in the area of cancer research.
Collaborators
Dr. Alex Saldanha, PhD, Computer Engineering, University of California Berkeley
Mr. Victor McGuire, CEO Renown, AI., Data Science/Machine Learning Engineer, B.S. United States Air Force Academy
