Timezone: »

 
Spotlight Poster
Proximity-Informed Calibration for Deep Neural Networks
Miao Xiong · Ailin Deng · Pang Wei Koh · Jiaying Wu · Shen Li · Jianqing Xu · Bryan Hooi

Thu Dec 14 03:00 PM -- 05:00 PM (PST) @ Great Hall & Hall B1+B2 #1618
in Poster Session 6 »
Confidence calibration is central to providing accurate and interpretable uncertainty estimates, especially under safety-critical scenarios. However, we find that existing calibration algorithms often overlook the issue of proximity bias, a phenomenon where models tend to be more overconfident in low proximity data (i.e., data lying in the sparse region of the data distribution) compared to high proximity samples, and thus suffer from inconsistent miscalibration across different proximity samples. We examine the problem over $504$ pretrained ImageNet models and observe that: 1) Proximity bias exists across a wide variety of model architectures and sizes; 2) Transformer-based models are relatively more susceptible to proximity bias than CNN-based models; 3) Proximity bias persists even after performing popular calibration algorithms like temperature scaling; 4) Models tend to overfit more heavily on low proximity samples than on high proximity samples. Motivated by the empirical findings, we propose ProCal, a plug-and-play algorithm with a theoretical guarantee to adjust sample confidence based on proximity. To further quantify the effectiveness of calibration algorithms in mitigating proximity bias, we introduce proximity-informed expected calibration error (PIECE) with theoretical analysis. We show that ProCal is effective in addressing proximity bias and improving calibration on balanced, long-tail, and distribution-shift settings under four metrics over various model architectures. We believe our findings on proximity bias will guide the development of fairer and better-calibrated} models, contributing to the broader pursuit of trustworthy AI.

Author Information

Miao Xiong (National University of Singapore)
Ailin Deng (National University of Singapore)
Pang Wei Koh (University of Washington)
Jiaying Wu (National University of Singapore)
Shen Li (national university of singaore, National University of Singapore)
Jianqing Xu (HIT)
Bryan Hooi (National University of Singapore)

More from the Same Authors