University of Waterloo Researchers Develop New AI Model to Reduce Bias and Enhance Trust in Machine Learning
Researchers from the University of Waterloo have created a groundbreaking explainable artificial intelligence (AI) model that aims to address bias and improve accuracy in machine learning-generated decisions and knowledge organization.
The Problem with Traditional Machine Learning Models
Traditional machine learning models often produce biased outcomes, favoring larger population groups or being influenced by unknown factors. Identifying these biases requires significant effort and can be challenging as patterns and sub-patterns come from different sources and classes.
In the medical field, biased machine learning results can have serious consequences. Hospital staff and medical professionals rely on datasets and computer algorithms to make critical decisions about patient care. However, these models may fail to detect rare symptomatic patterns in certain patient groups, leading to misdiagnoses and unequal healthcare outcomes.
The Solution: Pattern Discovery and Disentanglement (PDD)
Dr. Andrew Wong, a distinguished professor emeritus of systems design engineering at Waterloo, and his team have developed a new explainable AI model called Pattern Discovery and Disentanglement (PDD). This innovative model aims to untangle complex patterns from data and relate them to specific underlying causes, enabling more accurate and trustworthy decision-making.
By analyzing a large amount of protein binding data, Dr. Wong’s team demonstrated that entangled statistics can be disentangled to reveal the deep knowledge hidden within the data. This breakthrough led to the development of the PDD model, which has the potential to revolutionize pattern discovery in various industries.
The Value of PDD in Healthcare
Dr. Peiyuan Zhou, the lead researcher on Dr. Wong’s team, highlights the significance of PDD in bridging the gap between AI technology and human understanding. The PDD model can contribute to clinical decision-making and provide rigorous statistics and explainable patterns to support healthcare professionals in making reliable diagnoses and treatment recommendations for various diseases.
Through case studies, PDD has showcased its ability to predict medical outcomes based on patients’ clinical records and detect mislabels or anomalies in machine learning. This enables healthcare practitioners to provide better treatment recommendations and ensure more equitable healthcare outcomes.
The findings of this research have been published in the journal npj Digital Medicine. Moreover, the PDD model has received an NSER Idea-to-Innovation Grant of $125K and has garnered recognition in the industry. It is currently being commercialized through the Waterloo Commercialization Office.
With this groundbreaking AI model, the University of Waterloo researchers are taking significant strides towards reducing bias, enhancing trust, and improving accuracy in machine learning-generated decision-making and knowledge organization.