🤖 AI Summary
This work addresses the pressing need for efficient and reliable uncertainty quantification in deep neural networks deployed in safety-critical systems. To overcome limitations of existing stochastic regularization approaches—particularly the lack of a Bayesian interpretation for Stochastic Depth and insufficient validation in multi-task settings—the authors propose Monte Carlo Stochastic Depth (MCSD), reframing it as a Monte Carlo sampling mechanism to approximate Bayesian inference. The study establishes, for the first time, a theoretical connection between MCSD and variational inference. Comprehensive evaluations on object detectors such as YOLO and RT-DETR using COCO and COCO-O datasets demonstrate that MCSD achieves calibration (measured by ECE) and uncertainty ranking performance (via AUARC) slightly superior to Monte Carlo Dropout, while maintaining high mAP, robustness, and computational efficiency.
📝 Abstract
The deployment of deep neural networks in safety-critical systems necessitates reliable and efficient uncertainty quantification (UQ). A practical and widespread strategy for UQ is repurposing stochastic regularizers as scalable approximate Bayesian inference methods, such as Monte Carlo Dropout (MCD) and MC-DropBlock (MCDB). However, this paradigm remains under-explored for Stochastic Depth (SD), a regularizer integral to the residual-based backbones of most modern architectures. While prior work demonstrated its empirical promise for segmentation, a formal theoretical connection to Bayesian variational inference and a benchmark on complex, multi-task problems like object detection are missing. In this paper, we first provide theoretical insights connecting Monte Carlo Stochastic Depth (MCSD) to principled approximate variational inference. We then present the first comprehensive empirical benchmark of MCSD against MCD and MCDB on state-of-the-art detectors (YOLO, RT-DETR) using the COCO and COCO-O datasets. Our results position MCSD as a robust and computationally efficient method that achieves highly competitive predictive accuracy (mAP), notably yielding slight improvements in calibration (ECE) and uncertainty ranking (AUARC) compared to MCD. We thus establish MCSD as a theoretically-grounded and empirically-validated tool for efficient Bayesian approximation in modern deep learning.