Bayesian Optimization for Precision Agriculture with Scalable Probabilistic Models

Ruhana Azam, Sang T. Truong, samuel bonfim fernandes, Andrew D.B. Leakey, Alexander Lipka, Mohammed El-Kebir, Sanmi Koyejo

Keywords: bayesian optimization; genomic selection

Abstract: An overarching goal of precision agriculture is to select plants with desirable traits to boost crop yield and strengthen resilience to climate change and pest infestations. To select such crops, geneticists utilize predictive models to forecast plants' desirable traits from low-cost proxies. Although such data is abundant and easy to collect, identifying meaningful candidates is expensive. Bayesian optimization (BO) is a strong cost-effective solution to identify valuable proxies. In this work, we quantify the performance of BO with a collection of acquisition function and surrogate models for identifying good proxies, in a set of +4 million proxies. We find BO achieves comparable performance to random search while requiring significantly less computation. Despite traditional BO and random search techniques performing sufficiently well, both search techniques fail to leverage information from related tasks. To this end, we propose a pre-trained model as a transfer learning method. Using this benchmark, we conduct an extensive empirical study and demonstrate promising results on the transfer learning effect, paving a new way for an efficient optimization with a parsimonious sample size.

Submission Number: 8