“Multiclass Local Calibration with the Jensen–Shannon Distance” is on arXiv
“Multiclass Local Calibration with the Jensen–Shannon Distance” is now available on arXiv.
Abstract
Developing trustworthy Machine Learning (ML) models requires their predicted probabilities to be well-calibrated, meaning they should reflect true-class frequencies. Among calibration notions in multiclass classification, strong calibration is the most stringent, as it requires all predicted probabilities to be simultaneously calibrated across all classes. However, existing approaches to multiclass calibration lack a notion of distance among inputs, which makes them vulnerable to proximity bias: predictions in sparse regions of the feature space are systematically miscalibrated. In this work, we address this main shortcoming by introducing a local perspective on multiclass calibration. First, we formally define multiclass local calibration and establish its relationship with strong calibration. Second, we theoretically analyze the pitfalls of existing evaluation metrics when applied to multiclass local calibration. Third, we propose a practical method to enhance local calibration in Neural Networks, which enforces alignment between predicted probabilities and local estimates of class frequencies using the Jensen-Shannon distance. Finally, we empirically validate our approach against existing multiclass calibration techniques.
I am a PhD candidate in Artificial Intelligence at the University of Pisa, and I work with the Structured Machine Learning Group at the University of Trento.
My research focuses on uncertainty quantification and reliable machine learning, with particular interests in probabilistic calibration, conformal prediction, and human-AI decision making. More broadly, I am interested in developing the next generation of uncertainty-aware, trustworthy and human-centric AI frameworks.
I am always happy to chat about research, possible collaborations or interesting ideas, so feel free to reach out.