presentations

The growing popularity of electric vehicles (EVs) forces transportation agencies to consider the demand, and optimal location for public EV charging infrastructure. Most public charging stations are privately owned, and cluster near destinations or along highways. These trends, coupled with the convenience of at-home charging, has led to a scarcity of charging infrastructure in urban residential areas where potential users may not have a dedicated place to park a vehicle and charge. Ironically, urban drivers may be more inclined to use EVs compared to other drivers because of shorter commutes and lower range anxiety, but without a place to change at home, many forego adoption. For the second installment of the Spring 2026 eMERGE speaker series, Mike Montilla explores this issue, and presents recent research examining how the availability of public chargers, and policies promoting installation of public chargers within urban neighborhoods relate to EV adoption in dense cities with limited at-home parking.

During the COVID-19 pandemic, mobility networks derived from aggregated cell phone location data became a powerful tool for measuring population movement and inferring real-world contact patterns. In this talk, I will discuss my PhD research on mobility networks for pandemic response, describing parts of my job talk and job market experience. At the end, I’ll discuss some emerging directions from my new lab around generating mobility trajectories with generative AI models.

Data-intensive computing, exemplified by artificial intelligence (AI), is reshaping our lives, from healthcare to home security, but at a steep energy cost. In modern computing, most energy is spent moving data rather than processing it, creating an energy bottleneck known as the memory wall. Overcoming this challenge is crucial to enabling the energy-efficient electronics required for sustainable AI. Approaches such as analog in-memory computing and high-density, 3D on-chip memory offer the potential to overcome this bottleneck. However, realizing these approaches requires innovation in the fundamental building blocks of hardware, which, in turn, demands advances in device and materials engineering, including transport physics, interfaces, and integration. I will discuss how adaptive engineering leverages heterogeneous materials and devices to meet device-specific requirements and thus enables energy-efficient electronics. I will present key examples from my work. The first involves optimizing transport, interfacial, and thermal properties to enable adaptive oxide-based, low-power, non-volatile resistive random access memories for deep learning applications and exploit bias-induced phase transitions. The second leverages interface dipole engineering, monolithic 3D integration, and nanoscale process innovation for amorphous oxide semiconductor (AOS) two-transistor (2T) gain cell memory. Finally, I will conclude by outlining my vision, which combines adaptive engineering of novel materials and devices, 3D integration, and sustainable manufacturing to achieve unprecedented energy efficiency in computing hardware.

For decades, humanity operated under an illusion of abundance, exploiting oil, water, land and other planetary resources as if they were infinite. This paradigm is shifting as we confront resource scarcity. Automation technologies are deployed across critical infrastructure to optimize how energy grids, water networks, and transportation systems distribute finite resources efficiently.
In this talk, I will present three control perspectives and methods developed during my PhD that ensure efficiency, fairness, and welfare in closed-loop operation of large-scale infrastructure systems: (1) Game-theoretic MPC: An emerging control methodology modeling competitive self-interested agents with shared resources while incorporating predictions, dynamic models, and constraints into decision-making. I will present energy management results from a collaboration with a local DSO in Switzerland. (2) Welfarist control theory: A principled approach to aggregate incomparable heterogeneous cost functions, applied to water management. (3) Robust control techniques for maximal system efficiency in compute load allocations, developed with Google's Carbon Aware Computing team.

The global energy transition is transforming power systems through the rapid growth of renewables, data centers, and electrification across transportation, infrastructure, and industry. This talk highlights two critical challenges and our proposed solutions. First, distributed energy resources (DERs)—such as rooftop solar, home batteries, and electric vehicles—along with large new loads like data centers, require reliable and fair access to the distribution network. We introduce a dynamic operating envelope market, implemented through a forward auction, to allocate network injection and withdrawal rights. This mechanism enhances grid reliability while enabling equitable participation of diverse resources. Second, unlocking demand-side flexibility requires energy service providers to better understand customer behaviors and preferences in order to design sustainable and profitable business models. We develop AI agents powered by large language models that act as homo silicus for energy customers. These agents generate credible and diverse behavioral insights and capture responses to rare but high-impact events such as blackouts. Together, these approaches address renewable uncertainty, maintain competitive pricing, and advance demand-side participation in modern electricity markets.

Rapid growth of passenger electric vehicle (EV) adoption raises many challenges associated with increased electricity demand from charging. Limited electricity grid infrastructure capacity, expensive grid equipment costs, and emissions from electricity generation are all issues that EVs can exacerbate. However, EVs and other distributed energy resources (DERs), such as residential batteries, are flexible loads that can be managed to alleviate these challenges. This talk focuses on testing traditional management techniques for EVs and DERs to assess their efficacy given growing EV adoption. Specifically, we examine residential battery aggregation, vehicle-to-grid (V2G) charging economic feasibility, and managed EV charging for emissions reduction, and we propose solutions if traditional management methods fail. Together, this work helps accelerate near-term EV and V2G adoption with recommendations for best practices for EV and DER management.

Heavy-duty road transportation is among the most challenging sectors to decarbonize globally. Battery electric trucks (BETs) and fuel cell electric trucks (FCETs) have been proposed as key pathways toward zero-emission freight. However, technical and economic barriers continue to constrain their large-scale deployment, and the diverse usage scenarios of heavy-duty road transportation requires fleet-tailored solutions. New sources of real-world operational datasets (e.g., on-board sensing data, geospatial datasets) and computational methods enable us with more realistic assessments of truck decarbonization pathways. In this talk, I will present our recent observations on behavioral bottlenecks, evaluation and projection on technological feasibility, cost competitiveness, and decarbonization potential of low-carbon trucks, with a focus on practices across China, Europe, and the United States.

Technological advances have opened new avenues for designing market mechanisms in sociotechnical and urban infrastructure systems, from improving allocation efficiency through widespread data availability to enabling real-time algorithm implementation. While these advances hold significant promise, they also introduce challenges around equity, privacy, data uncertainty, and security that traditional mechanisms often fail to address. My research develops data-driven and online learning algorithms and incentive schemes to tackle these challenges, advancing the science and practice of market design for socially sustainable resource allocation in large-scale sociotechnical and urban infrastructure systems. In this talk, I will highlight my work on designing approximation algorithms for NP-hard problems that emerge when incorporating fairness and security considerations into smart mobility applications. Specifically, I will discuss new approaches for incorporating fairness considerations in congestion pricing and scalable algorithms for multi-resource Bayesian Stackelberg security games, illustrating both the theoretical contributions and their applications in traffic management and parking enforcement.

Traffic systems are rapidly evolving toward mixed autonomy, where human-driven and automated vehicles (AVs) coexist. Achieving safe, efficient, and sustainable mobility in this setting requires algorithms that combine the strengths of traditional models, the flexibility of machine learning, and the practicality of control design and experimentation.
In this talk, I will present my research on physics-informed learning and control for mixed autonomy traffic systems. I begin with the CIRCLES project, where we showed that carefully designed control algorithms for a small fraction of automated vehicles can dissipate stop-and-go traffic waves and reduce fuel consumption. These results were validated in the MegaVanderTest, a large-scale field deployment of 100 AVs on a busy section of I-24 during the morning rush. This motivates the need for accurate, scalable models of traffic dynamics to guide such strategies. At the modeling level, I will introduce the Neural Finite Volume Method (NFVM), a physics-informed machine learning architecture for hyperbolic PDEs that embeds conservation laws, achieving up to an order of magnitude lower error than classical schemes such as Godunov’s method, outperforming ENO/WENO, and rivaling discontinuous Galerkin solvers at far lower complexity. Additionally, I will highlight how this research extends toward a broader vision: combining modeling, control, and real-world experimentation to advance intelligent mobility systems.

Modern Artificial Intelligence (AI) and Machine Learning (ML) algorithms can deliver significant performance improvements for decision-making under uncertainty, where traditional, worst-case algorithms are often too conservative. These improvements can potentially be transformative for energy and sustainability applications, where rapid advances are needed to facilitate the energy transition. However, AI and ML typically lack provable performance guarantees, hindering their deployment to real-world problems in energy and sustainability where safety and reliability are critical. In this talk, I will discuss recent work developing algorithms that bridge the gap between the strong average-case performance of AI/ML and the worst-case guarantees of traditional algorithms. In particular, I will focus on the question of how to robustly leverage the recommendations of black-box AI “advice” for general classes of online optimization problems, describing both algorithmic upper bounds and fundamental limits on performance in these settings. In the second half of the talk, I will move beyond this black-box approach, highlighting recent steps toward developing new risk- and reliability-aware methods for ML training to enable better decision-making performance. Throughout, I will emphasize applications of these frameworks to energy and sustainability-related problems such as energy resource management and sustainable computing.

Clean energy transitions in many jurisdictions involve dramatic increases in shares of variable wind and solar power in their electricity grids, supplemented by clean firm power to provide reliability. To plan a resilient, clean energy future, we will need to account for natural resource constraints as we develop and deploy new technologies. To address this aim, I blend perspectives from earth science, energy system modeling, and energy economics. I will discuss my work that incorporates multi-decadal wind and solar weather data into energy system models, showing that it is critical to incorporate this variability for evaluating seasonal and interannual benefits of energy storage. Long-duration energy storage can make reliable wind-solar-battery electricity systems more affordable. Geologic hydrogen storage, thermal energy storage in dirt, and metal-air batteries are promising examples of this. In additional work, I estimate carbon abatement costs of building heat electrification constrained by temperature data at the U.S. census tract level. This talk illustrates that wind and solar constraints guide energy storage opportunities, and temperature constraints guide heat pump opportunities, informing energy infrastructure investment and research priorities.

Tracking carbon emissions in near-real-time is essential for informing timely climate action and evaluating the effectiveness of mitigation policies. In this talk, Dr. Dou will introduce her work on developing the world’s first high-resolution, near-real-time datasets of CO2 and CH4 emissions by integrating satellite observations, ground-based activity data, and machine learning. The innovative methods are adopted in multiple international collaborative projects such as Carbon Monitor and GRACED, providing daily CO2 and CH4 emission estimates across multiple sectors for global coverage. Dr. Dou will also share how these data have been used to assess the immediate impact of socio-economic changes, such as the pandemic or energy crises, on emissions. Moving forward, this research aspires to shed light on the dance between the energy transition and GHG emissions, offering insights to guide future policy decisions and technological advancements.

Cities are complex systems shaped by rich patterns across space and time. Understanding and predicting these patterns requires more than accurate models—it requires trustworthy spatial intelligence. In this talk, Dingyi Zhuang will present his research agenda along three complementary directions. First, he develops spatiotemporal and multimodal modeling frameworks with graph neural networks and tensor learning to capture and predict correlations across space and time. Second, he designs uncertainty quantification and calibration techniques to ensure that the relationships learned by these models are reliable, interpretable, and fair. Third, he extends toward spatial reasoning, leveraging large language models and vision-language models to embed physical constraints and social norms into learned structures, advancing toward richer world models of cities. Together, these directions outline a coherent vision: from learning relationships in spatiotemporal data, to validating their trustworthiness, to reasoning about the rules that govern urban environments.