Sean Kudrna

Contributors: Zohair Hashmi, Sajal Chandra, Rayaan Siddiqi, Sean Kudrna
Github Repository: https://github.com/zohairhashmi/uncertainty-quantification

This project explores the complex interplay between algorithmic fairness and the uncertainties inherent in machine learning models. Focusing on aleatoric (related to data) and epistemic (related to model knowledge) uncertainties, our research developed methods to quantify these uncertainties with the objective of improving the fairness… Show more

Contributors: Zohair Hashmi, Sajal Chandra, Rayaan Siddiqi, Sean Kudrna
Github Repository: https://github.com/zohairhashmi/uncertainty-quantification

This project explores the complex interplay between algorithmic fairness and the uncertainties inherent in machine learning models. Focusing on aleatoric (related to data) and epistemic (related to model knowledge) uncertainties, our research developed methods to quantify these uncertainties with the objective of improving the fairness and accuracy of decisions made by machine learning algorithms.

Objective:
Our research aimed to address the challenge of enhancing fairness in machine learning models by mitigating uncertainties through sophisticated characterization techniques. This work contributes to making algorithms more transparent and equitable.

Approach:
We utilized state-of-the-art metrics and tools, such as the AIF360 Fairness Metric and Keras_Uncertainty, to assess and quantify fairness and uncertainties within machine learning models. The project also involved experimenting with deep ensemble learning models and applying dropout techniques to evaluate the effect of uncertainties under various conditions.

Findings:
Our research highlighted the significant role of aleatoric and epistemic uncertainties in affecting the fairness metrics of machine learning models, presenting challenges in achieving fairness. Our findings display the need for more advanced methods in uncertainty quantification.

Significance:
The implications of our work extend across several fields, including autonomous driving, healthcare, and criminal justice, where the fairness of machine learning models is crucial. Our results add to the ongoing dialogue on ethical AI and the development of more accountable ML practices.

Future Work:
This project sets the stage for further research into optimizing the balance between fairness and uncertainty in machine learning, suggesting potential for impactful applications across a variety of sectors. Show less