B.5.1 Room
Organizer: Cerchiello
Chair: Cerchiello
Pascual, J. D. Fair Overlap Number of Balls (Fair-ONB): A Data-Morphology-based Undersampling Method for Bias Reduction
Abstract
| One of the key issues regarding classification problems in Trustworthy Artificial Intelligence is ensuring Fairness in the prediction of different classes when protected (sensitive) features are present. Data quality is critical in these cases, as biases in training data can be reflected in machine learning, impacting human lives and failing to comply with current regulations. One strategy to improve data quality and avoid these problems is preprocessing the dataset. Instance selection via undersampling can foster balanced learning of classes and protected feature values. Performing undersampling in class overlap areas close to the decision boundary should bolster the impact on the classifier. This work proposes Fair Overlap Number of Balls (Fair-ONB), an undersampling method that harnesses the data morphology of the different data groups (obtained from the combination of classes and protected feature values) to perform guided undersampling in overlap areas. It employs attributes of the ball coverage of the groups, such as the radius, number of covered instances and density, to select the most suitable areas for undersampling and reduce bias. Results show that the Fair-ONB method improves model Fairness with low impact on the classifier’s predictive performance. |
Giudici, P. Safe Artificial Intelligence
Abstract
| The rapid development of artificial intelligence applications based on machine learning is creating not only many opportunities but also risks. The recent regulatory and political debate, at the international level, emphasizes the urgent need to develop appropriate statistical methods that can measure the safety and the risks of AI applications. In line with this emerging need, we define which metrics to develop and an example of a recent proposal, along with its advantages and disadvantages. |
Beraha, M. On the Differential Privacy of Bayesian Nonparametric Posteriors
Abstract
| We propose a novel data release mechanism for privacy-preserving synthetic data generation, based on posterior predictive sampling from the Pitman–Yor (PY) process. This method satisfies a notion of differential privacy under suitable conditions on model parameters and the amount of synthetic data released. Notably, it preserves essential statistical features of the confidential dataset while ensuring robust privacy protection. We present the main theoretical result characterizing the privacy guarantees, and discuss how the released synthetic data remain informativein both frequentist and Bayesian senses. |