INFORMS 2021 Program Book

INFORMS Anaheim 2021

MB19

2 - Racial Equity in Energy and Sustainability: A Case Study in Mexico Rodrigo Mercado Fernandez, Appalachian State University, Richmond, VA, 23220, United States, Erin Baker Using Mexico as a case study, we employ a bottom-up model of the electrical system to identify critical geographic areas of investment for installed capacity and transmission that are robust across a set of climate mitigation pathways derived from multiple Integrated Assessment Models. We find that more diverse energy portfolios are associated with relatively less transmission investment; and that despite a lack of robustness in the location of installed capacity investments, investment in transmission expansion is fairly robust across pathways. 3 - Exploring the Role of Electric Vehicles in Africa’s Energy Transition Michael Dioha, Carnegie Institution for Science, Stanford, CA, United States, Lei Duan, Tyler Ruggles, Sara Bellocchi, Ken Caldeira We employ a bottom-up modelling framework to examine the interplay of electric vehicles (EVs) and variable renewables (VRE) in Africa using Nigeria as a case study. Our results indicate that despite having a natural gas-dominated electricity system, the deployment of EVs can support the decarbonization of the Nigerian transport sector but at a relatively high cost. The cost of EVs would need to drop by ~40% to become cost-competitive. However, if VRE delivers the EVs power requirement with a bidirectional smart charging strategy, then the cost of EVs would need to decline by only ~30% to be a cost-effective option. Not all EVs need to participate in a bidirectional charging strategy in order to realize its full benefits; there is substantial benefit from flexibility in charging loads. Robust policies are needed to support EVs. MB19 CC Room 203A In Person: Data Mining Contributed Session Chair: Anna Khalemsky, Maale Adumim, 9856227, Israel 1 - Random Forest Incorporated Multi-fidelity Cokriging to Handle the Nonstationarity in the Data Mithun Ghosh, University of Arizona, Tucson, AZ, United States, Qiang Zhou We propose an ensemble framework to model the multi-fidelity data with heterogeneity across the input space. Computationally, modeling a p-level recursive multi-fidelity-based cokriging is equivalent to build p individual krigings. We use the recursive cokriging model in the random forest framework, where the high-fidelity data with the bootstrapped samples are split into homogenous sections. We propose an efficient recursive partitioning of the tree that enables the model to handle the nonstationarity by using a stationary covariance function. The dependency between trees in the random forest makes the prediction distribution of our model using the Gaussian Mixture model. 2 - Efficient and Trusted Resource Allocation by Enhancing Theory Novel methodology for capturing causal impact heterogeneity for highly efficient targeting using large scale observational data. Supports justifiable resource allocation with interpretable models that enable theory-driven qualitative constraints and incorporate important findings of causal machine learning. Quantifies the interpretability performance tradeoff with untapped targeting efficiency.Applied on targeting resources to increase physical activity levels in the Turkish adult population, we find that imposing literature driven qualitative constraints, and model modification based on causal machine learning with XAI both increase the expected public health benefits. 3 - A Decision Support Tool to Estimate Rank in Plant Breeding Experiments Reyhaneh Bijari, PhD Student, Iowa State University, Ames, IA, United States, Hanisha Vemireddy, Sigurdur Olafsson Plant breeders aim to select the genotypes with the best genetic properties. This decision may fall in ranking these genotypes based on some phenotypic traits. However, the uncertainty of the ranking makes this decision hard as it is only possible to observe each genotype in a few environments. Each genotype has significant genetics-by-environment effects, resulting in one genotype appearing better than its genetic potential or the other way around. We propose a bootstrapping approach to construct confidence intervals around rank, capturing its inherent uncertainty. We show that the tool is effective as the empirical coverage of the confidence interval closely matches the theoretical coverage. Driven Models with Causal Machine Learning Ozden Gur Ali, Koc University, Istanbul, Turkey

MB15 CC Room 201C In Person: Advances in Data Analytics and Applications General Session Chair: Tianyi Lin, University of California, Berkeley, Berkeley, CA, 94720-2502, United States 1 - Computing Wasserstein Barycenters: Easy or Hard? Jason Altschuler, Massachusetts Institute of Technology, Cambridge, MA, United States Averaging data distributions is a core subroutine throughout data science. Wasserstein barycenters (a.k.a. Optimal Transport barycenters) provide a natural approach for this problem that captures the geometry of the data, and are central to diverse applications in machine learning, statistics, and computer graphics. Despite considerable attention, it remained unknown whether Wasserstein barycenters can be computed in polynomial time. Our recent work provides a complete answer to this question and reveals a surprising “curse of dimensionality”.Joint work with Enric Boix 2 - Optimal Transport for Text Mining Zhiyue Hu, University of California-Berkeley, Berkeley, CA, United States Topic sparsity refers to the observation that individual documents usually focus on several salient topics instead of covering a wide variety of topics, and a real topic adopts a narrow range of terms instead of a wide coverage of the vocabulary. Understanding this topic sparsity is especially important for analyzing user- generated web content and social media, which are featured in the form of extremely short posts and discussions. As topic sparsity of individual documents in online social media increases, so does the difficulty of analyzing the online text sources using traditional methods. In this paper, we propose two novel neural models by providing sparse posterior distributions over topics based on the Gaussian sparsemax construction, enabling efficient training by stochastic back- propagation. 3 - An Empirical Investigation of Factors Influencing Performance of Decentralized Applications Luv Sharma, University of South Carolina, Columbia, SC, 29206, United States, Moonwon Chung, Jie Lian We investigate factors impacting the performance of decentralized application (DApp) that are built on the Ethereum blockchain. 4 - Characterizing and Comparing Covid-19 Misinformation Across Languages, Countries and Platforms Jacqueline Otala, Clarkson University, Potsdam, NY, United States, Golshan Madraki, Isabella Grasso, Yu Liu, Jeanna Matthews We investigate COVID-19 misinformation in multiple languages/countries: Chinese/China, English/USA, and Farsi/Iran; on multiple platforms: Twitter, Facebook, Instagram, WhatsApp, Weibo, WeChat and TikTok. Utilizing opportunistic sampling, we compiled 200 items of viral and debunked misinformation across these languages, countries and platforms from January 1- August 31 2020. While it was observed that COVID-19 misinformation on social media varied across different languages, politics was observed as the root of most collected misinformation across all three languages. We further observe the impact of government platform restrictions on content in China, Iran, and USA. MB18 CC Room 202B In Person: Macro Energy Systems: Energy Infrastructure Resilience General Session Chair: Benjamin D Leibowicz, University of Texas-Austin, Austin, TX, 78712-1591, United States 1 - Disaster Resilience Planning under Uncertainty: A Nexus Approach Rachel Moglen, University of Texas at Austin, Austin, TX, United States, Benjamin D. Leibowicz Natural disasters pose a serious threat to our infrastructure systems, inflicting significant losses of property and life annually. Decision-makers aim to prevent these negative consequences by hardening infrastructure, a process complicated by the inter-dependencies between critical infrastructure systems. We develop and implement a two-stage stochastic program to capture the uncertainty in disaster realization, minimizing expected unmet water and power demand across disasters. Our results show that our proposed nexus optimization approach performs significantly better than if the water and power network were optimized independently. The results also show an emphasis on power system hardening, even when water service is strictly prioritized, due to the dependency of the water system on power for drinking water treatment.

49