Search for: All records

Human development is a threat to biodiversity and conservation organizations (COs) are purchasing land to protect areas for biodiversity preservation. COs have limited budgets and cannot purchase all the land necessary to perfectly preserve biodiversity, and human activities are uncertain, so exact developments are unpredictable. We propose a multistage, robust optimization problem with a data-driven hierarchical-structured uncertainty set which captures the endogenous nature of the binary (0-1) human land use uncertain parameters to help COs choose land parcels to purchase to minimize the worst-case human impact on biodiversity. In the proposed approach, the problem is formulated as a game between COs, which choose parcels to protect with limited budgets, and the human development, which will maximize the loss to the unprotected parcels. We leverage the cellular automata model to simulate the development based on climate data, land characteristics, and human land use data. We use the simulation to build data-driven uncertainty sets. We demonstrate that an equivalent formulation of the problem can be obtained that presents exogenous uncertainty only and where uncertain parameters only appear in the objective. We leverage this reformulation to propose a conservative $K$-adaptability reformulation of our problem that can be solved routinely by off-the-shelf solvers like Gurobi or CPLEX. The numerical results based on real data show that the proposed method reduces conservation loss by 19.46% on average compared to standard approaches used in practice for biodiversity conservation. 

We consider the parameter estimation problem of a probabilistic generative model prescribed using a natural exponential family of distributions. For this problem, the typical maximum likelihood estimator usually overfits under limited training sample size, is sensitive to noise and may perform poorly on downstream predictive tasks. To mitigate these issues, we propose a distributionally robust maximum likelihood estimator that minimizes the worst-case expected log-loss uniformly over a parametric Kullback-Leibler ball around a parametric nominal distribution. Leveraging the analytical expression of the Kullback-Leibler divergence between two distributions in the same natural exponential family, we show that the min-max estimation problem is tractable in a broad setting, including the robust training of generalized linear models. Our novel robust estimator also enjoys statistical consistency and delivers promising empirical results in both regression and classification tasks.