I have started to serve as Associate Editor of IEEE Transactions on Reliability!
IEEE Transactions on Reliability is a leading peer-reviewed quarterly journal published by the IEEE Reliability Society. It focuses on quantitative methods, engineering theory, and practical applications that predict, measure, and improve the safety, maintainability, and dependability of hardware, software, and complex systems. Please find the editorial board of IEEE Transactions on Reliabilty below:
Dr. Xiaoge Zhang delivered a talk on “Uncertainty quantification for reliable AI and its applications to healthcare and traffic systems” at Tongji University, China
As artificial intelligence (AI) is increasingly integrated into industrial applications, ensuring its reliability has become imperative, particularly in high-stakes decision-making settings. In this talk, I will present a general uncertainty quantification (UQ)-based framework for implementing reliable AI that proactively addresses both “known unknowns” and “unknown unknowns” that AI models inevitably encounter after deployment. Our proposed framework leverages the complementarity of model-based UQ methods (e.g., Monte Carlo dropout, variational inference) and model-agnostic, distribution-free conformal prediction (CP), together enabling AI models to explicitly express “I do not know” in situations where they are uncertain. In the UQ-based framework, model-based UQ methods are used to manage “unknown unknowns” by detecting inputs that lie outside the applicability domain (AD) of the trained models, thereby defining a region in which the assumption of data exchangeability is more plausible for conformal prediction. Within the model’s AD, conformal prediction is used to manage “known unknowns”, and provides finite-sample, distribution-free guarantees of model reliability by constructing prediction intervals that contain the true values with a user-specified coverage probability. I will use two case studies to demonstrate the practical utility of the proposed framework in supporting accurate and reliable pathology-based cancer diagnosis and traffic prediction over large-scale road networks.
Research paper accepted by IEEE Transactions on Reliability
In this paper, we develop a data-efficient learning-based approach for accurately estimating the time-varying kinematic reliability of mechanisms. To improve data efficiency, we develop an adaptive sampling scheme that is guided by the prediction uncertainty of the resultant deep learning model. Overall, the proposed methodology follows a four-step procedure. First, a spectral-normalized neural Gaussian process (SNGP) is employed to quantify the prediction uncertainty of the deep learning model used to evaluate the time-varying kinematic reliability of mechanisms. Unlike other uncertainty quantification methods, SNGP has two unique characteristics: SNGP normalizes the weights of hidden layers of neural network to preserve the relative distances among samples during data transformation; SNGP replaces the conventional dense output layer of neural network with a random Fourier features-based Laplace approximation of Gaussian process for distance-aware uncertainty quantification. Next, based upon the uncertainty estimated by SNGP, we devise an active learning function to adaptively identify and select the most valuable next data sample to add to the training set. Thirdly, based on a predefined data acquisition metric, we iteratively add data points that maximize the active learning function to the training set and continue updating the deep learning model until a termination condition is met. Finally, we benchmark the developed approach for time-varying reliability analysis of mechanisms against Monte Carlo simulation, PHI2, and the Monte Carlo dropout-based uncertainty quantification method in terms of estimation accuracy and computational efficiency using three numerical examples. Computational results suggest that the developed method substantially reduces the number of performance function evaluations while achieving reliability estimates that are more accurate than those from MC dropout and comparable to those from MCS and PHI2.
