Xiao Liu

In this article, novel gradient-based online learning algorithms are developed to investigate an important environmental application: real-time river pollution source identification, which aims at estimating the released mass, location, and time of a river pollution source based on downstream sensor data monitoring the pollution concentration. The pollution is assumed to be instantaneously released once. The problem can be formulated as a non-convex loss minimization problem in statistical learning, and our online algorithms have vectorized and adaptive step sizes to ensure high estimation accuracy in three dimensions which have different magnitudes. In order to keep the algorithm from sticking in the saddle points of non-convex loss, the “escaping from saddle points” module and multi-start setting are derived to further improve the estimation accuracy by searching for the global minimizer of the loss functions. This can be shown theoretically and experimentally as the O(N) local regret of the algorithms and the high probability cumulative regret bound O(N) under a particular error bound condition in loss functions. A real-life river pollution source identification example shows the superior performance of our algorithms compared with existing methods in terms of estimation accuracy. Managerial insights for the decision maker to use the algorithms are also provided.

Intermittent demand occurs commonly for spare parts in the heavy-duty vehicle industry. Demand uncertainty and intermittency pose challenges to demand forecasting by conventional models. Support vector machine (SVM) models have been observed to yield competitive accuracy with existing models. However, there are still limitations for basic SVM models. First, the time-consuming computation does not bring a statistically significant accuracy improvement. Second, the forecasting-based inventory performance has not been sufficiently explored. Third, scarce explanations of model robustness are offered for demand forecasting. We build an adaptive univariate SVM (AUSVM) model to forecast intermittent demand. Its effectiveness, compared to 12 existing models and an improved neural-network, is demonstrated by real-world data from a heavy-duty vehicle spare-part company. AUSVM has an apparent advantage in computation time over basic SVM and neural networks. The computational results of the heavy-duty vehicle case indicate that, compared to well-known parametric models, AUSVM achieves a statistically significant accuracy improvement and better inventory performance for the group of non-smooth demand series. Discussions are presented on why AUSVM works for demand forecasting and inventory control of heavy-duty vehicle spare parts. Several insights are revealed for practitioners in the heavy-duty vehicle industry.

We address a problem of preventing an interdiction on water supply networks by building a Bayesian Stackelberg game model involving stakeholders of a defender and an interdictor. The defender initiates to allocate resource to network components to make a trade-off between network resilience measured by water satisfaction rate and the defender's cost, whereas the interdictor follows to interdict a component with the objectives of maximising the destruction level on the network structure and minimising the interdictor's cost. Specifically, the defender adopts mixed defence strategies, which implies that the interdictor is uncertain of the defender's resource allocation. Moreover, we propose sufficient conditions for the elimination of the dominated defence and interdiction strategies. A decomposed iterative learning algorithm (DILA) and a smallest-depth binary-partition based hierarchical algorithm (SBHA) are developed to reduce the sizes of the defence and interdiction strategy sets, respectively, thus analysing the optimal mixed defence strategies. Finally, a real case study with private information is conducted, thus providing valuable suggestions for the defender's resource allocation against interdictions.

YouTube presents an unprecedented opportunity to explore how machine learning methods can improve healthcare information dissemination. We propose an interdisciplinary lens that synthesizes machine learning methods with healthcare informatics themes to address the critical issue of developing a scalable algorithmic solution to evaluate videos from a health literacy and patient education perspective. We develop a deep learning method to understand the level of medical knowledge encoded in YouTube videos. Preliminary results suggest that we can extract medical knowledge from YouTube videos and classify videos according to the embedded knowledge with satisfying performance. Deep learning methods show great promise in knowledge extraction, natural language understanding, and image classification, especially in an era of patient-centric care and precision medicine.