Proceedings Abstracts of the Twenty-Third International Joint Conference on Artificial Intelligence

Deep Sparse Coding Based Recursive Disaggregation Model for Water Conservation / 2804
Haili Dong, Bingsheng Wang, Chang-Tien Lu

The increasing demands on drinkable water, along with population growth, water-intensive agriculture and economic development, pose critical challenges to water sustainability. New techniques to long-term water conservation that incorporate principles of sustainability are expected. Recent studies have shown that providing customers with usage information of fixtures could help them save a considerable amount of water. Existing disaggregation techniques focus on learning consumption patterns for individual devices. Little attention has been given to the hierarchical decomposition structure of the aggregated consumption. In this paper, a Deep Sparse Coding based Recursive Disaggregation Model (DSCRDM) is proposed for water conservation. We design a recursive decomposition structure to perform the disaggregation task, and introduce sequential set to capture its characteristics. An efficient and effective algorithm deep sparse coding is developed to automatically learn the disaggregation architecture, along with discriminative and reconstruction dictionaries for each layer. We demonstrated that our proposed approach significantly improved the performance of the benchmark methods on a large scale disaggregation task and illustrated how our model could provide practical feedbacks to customers for water conservation.