Knowledge engineering principles for the effective management of complex systems

点击翻译按钮获取中文摘要

This research focuses on investigating advanced machine learning techniques with applications to engineering and business management. In the engineering area, we look at the analysis of the risk of failure of drinking water facilities. Contrary to what one may think, waterborne disease outbreaks in developed countries seem to continue to occur on a relatively regular basis. The reason for this phenomenon is not the lack of adequate treatment technologies. Rather it is the cumulative impact of a sequence of small individual events, each of which is relatively harmless on its own, that leads to the outbreak - hence the systemic nature of the causes of the failures, and the relevance of machine learning techniques such as kernels and pattern recognition algorithms (for the classification of a sequence of events as 'collapse-oriented' or 'survival-oriented'). In the business management area, the focus is on the topic of sales forecasting and management, which has a direct connection with the issue of job losses in the manufacturing sector in Ontario. This type of job losses is a direct consequence of the drop in manufacturing sales, which is itself due to the appreciation of the Canadian dollar against the United States dollar. Such an appreciation leads indeed to the odd situation where Canadian products suddenly become more expensive for American customers, while the value-adding features of these very same products have not changed at all - hence the drop in sales. Laying-off workers has so far been the quickest way, for manufacturers, to react to this drop in sales. There is, however, no evidence that the laying-off of workers is the optimal control action (in the control-theoretic sense of the term) to react to currency appreciation. The applicant and his co-workers developed a quantified system to help salespeople and managers analyze their interactions with customers, develop sales forecasts and design adequate action plans for the future. In this project, we intend to investigate both of these applications using the state-of-the-art techniques for machine learning, knowledge representation and pattern analysis.