Dynamic Resource Allocation in Conservation （动态资源分配保护）.pdf

Proceedings of the Twenty-Fifth AAAI Conference on Artificial Intelligence Dynamic Resource Allocation in Conservation Planning Daniel Golovin Andreas Krause Beth Gardner Sarah J. Converse Steve Morey ¨ Caltech ETH Zurich NCSU USGS Patuxent WRC USFWS Pasadena, CA, USA Zurich, Switzerland Raleigh, NC, USA Laurel, MD, USA Portland OR, USA Abstract clairvoyant policy with knowledge of the future availability of resources. While there is a significant amount of both Consider the problem of protecting endangered species by theoretical and applied work on conservation planning (which selecting patches of land to be used for conservation purposes. we review in §5), we are unaware of principled approaches Typically, the availability of patches changes over time, and recommendations must be made dynamically. This is a chal- that can solve such dynamic problems on a realistic scale. lenging prototypical example of a sequential optimization We further conduct a detailed computational study, evalu- problem under uncertainty in computational sustainability. Ex- ating our approach on the problem of conservation planning isting techniques do not scale to problems of realistic size. In for three rare taxa (see Fig. 1(a)) in the Pacific Northwest of this paper, we develop an efficient algorithm for adaptively the United States. Our case study uses complex patch dy- making