About

My research interests primarily focus on robustness in machine learning, such as adversarial robustness, out-of-distribution generalization, and fairness. In addition, I am highly passionate about developing reliable AI-driven models tailored for healthcare, with a particular focus on diabetes management. In my lab, we explore various research themes for designing robust machine learning models in healthcare.

Research topics

• Economics
• Environmental science
• Computer Science
• Engineering
• Environmental economics
• Waste management
• Environmental engineering
• Food science
• Agricultural economics
• Chemistry
• Agricultural science
• Water resource management
• Environmental protection
• Geography
• Agronomy

Selected publications

• NPC type three-level PV grid-connected inverter S-FCS-MPC control strategy

  Research Square · 2024-11-14 · 1 citations

  preprintOpen accessSenior author
  PublisherOA PDFDOI

• Modeling of Electrowinning Process in Zinc Hydrometallurgy Based on COMSOL

  2022-07-14

  article

  This paper aims at the problem of the influence of the gas evolved from the electrode on the performance of the electrolytic cell during the electrowinning process of zinc hydrometallurgy. Based on the COMSOL Multiphysics software, a visually intuitive coupling model of the distribution of the tertiary current and the flow of bubbles in the zinc electrolysis cell is established. The effects of the gas released from the electrode on the concentration distribution of zinc ions, current efficiency and power consumption in the electrolytic cell were studied. These works provide an important basis for further improvement of the structure and operation of zinc electrolytic cells and subsequent optimization of electrolytic cell performance.

  PublisherDOI

• Optimal allocation of tomato supply to minimize greenhouse gas emissions in major U.S. metropolitan markets

  Resources Conservation and Recycling · 2022-09-19 · 6 citations

  articleCorresponding
  PublisherDOI

• Non-linearity in Marginal LCA: Application of a Spatial Optimization Model

  Frontiers in Sustainability · 2021-05-05 · 9 citations

  articleOpen access1st author

  Typical applications of LCA assume that the magnitude of life-cycle impact grows proportionally to the volume of demand, while in reality the additional impact due to marginal increase in demand may differ from the average impact. In the literature, the calculation of marginal life-cycle impacts often involves the use of optimization models, where typically the total economic costs are minimized. However, modeling spatially explicit marginal responses of a system involving multiple producers and consumers has not been discussed in LCA literature. In this paper, we demonstrate a spatial optimization technique for modeling marginal responses of a multi-producer, multi-consumer system. Our model determines the optimal production-by-location mix and associated environmental stressor at minimum systems cost. We demonstrate the model using a preliminary case study on blue water consumption by potato. We collected state-by-state data on potato yield, cost of potato production, and water use for irrigation, as well as interstate transportation fuel costs. We also estimated the marginal increase in demand for potato following USDA's recommended diet. The results show that the cradle-to-gate blue water consumption of potatoes based on 2016 demand was 96 m 3 /ton potato, which changes non-linearly along with the growth of potato demands. In order to meet the USDA's recommended diet, the additional demand on potato (530,000 ton per year) would result in a 29% lower blue water consumption per ton of potato (68 m 3 /ton) as compared to the average result of the current production system. In addition, we tested the model to analyze the marginal impacts under two scenarios: (1) high fuel tax and (2) high water price. The preliminary results indicate that water pricing is more effective than a fuel tax increase in reducing the marginal blue water consumption of potato based on our scenarios of the recommended diet demand. The results demonstrate that our model can be used to understand the non-linear behavior of marginal effect over demand crease, and for testing alternative policy scenarios involving a system with multiple producers and consumers across regions.

  PublisherOA PDFDOI

• Method to decompose uncertainties in LCA results into contributing factors

  The International Journal of Life Cycle Assessment · 2021-04-20 · 23 citations

  articleOpen access1st authorCorresponding
  PublisherOA PDFDOI

• Optimal Allocation of Tomato Supply to Minimize Greenhouse Gas Emissions in Major U.S. Metropolitan Markets

  Research Square · 2021-01-13 · 1 citations

  preprintOpen access

  Abstract The United States food system requires energy, water, and land in significant proportions, releases large quantities of greenhouse gases, and contributes to other environmental concerns. Meeting future demand for fresh food will be especially challenging, requiring the adoption of holistic, systems-level thinking to maximize production and supply while limiting consequences to the climate and natural resources. We have developed a cradle-to-market life-cycle environmental model to assess the greenhouse gas footprint of fresh tomatoes supplied to ten of the largest metropolitan statistical areas in the United States. A linear optimization algorithm is applied to determine the optimal tomato distribution scheme that will minimize tomato-related greenhouse gas emissions across all ten areas. Monte Carlo simulation was performed to assess the uncertainties in the data. Results indicate that the current tomato distribution scheme is suboptimal; re-allocating the fresh tomato supply across these ten areas has the potential to decrease transportation-related emissions by 34% and overall tomato-related greenhouse gas emissions by 13%—from 277,000 MTCO 2 e to 242,000 MTCO 2 e. The substantial variability of the optimized scenario raises questions about its practical implementation. Ultimately, however, production practices and geographic conditions (such as soil and climate) are more significant with respect to environmental impact than the supply allocation or the seasonality of supply. Our analysis found a roughly six-fold difference between Philadelphia tomatoes sourced from open-field Virginian production (0.38 kgCO 2 e/kg) compared with controlled-environment Mexican production (2.3 kgCO 2 e/kg).

  PublisherOA PDFDOI

• Accelerating the pace of ecotoxicological assessment using artificial intelligence

  AMBIO · 2021-08-24 · 23 citations

  articleOpen access

  Abstract Species Sensitivity Distribution (SSD) is a key metric for understanding the potential ecotoxicological impacts of chemicals. However, SSDs have been developed to estimate for only handful of chemicals due to the scarcity of experimental toxicity data. Here we present a novel approach to expand the chemical coverage of SSDs using Artificial Neural Network (ANN). We collected over 2000 experimental toxicity data in Lethal Concentration 50 (LC50) for 8 aquatic species and trained an ANN model for each of the 8 aquatic species based on molecular structure. The R 2 values of resulting ANN models range from 0.54 to 0.75 (median R 2 = 0.69). We applied the predicted LC50 values to fit SSD curves using bootstrapping method, generating SSDs for 8424 chemicals in the ToX21 database. The dataset is expected to serve as a screening-level reference SSD database for understanding potential ecotoxicological impacts of chemicals.

  PublisherOA PDFDOI

• What contributes more to life-cycle greenhouse gas emissions of farm produce: Production, transportation, packaging, or food loss?

  Resources Conservation and Recycling · 2021 · 44 citations

  1st authorCorresponding
  • Environmental science
  • Agricultural science
  • Waste management
  PublisherDOI

• Model Predictive Control for the Process of MEA Absorption of CO2 Based on the Data Identification Model

  Processes · 2021-01-19 · 14 citations

  articleOpen access

  The absorption process of CO2 by ethanolamine solution is essentially a dynamic system, which is greatly affected by the power plant startup and flue gas load changes. Hence, studying the optimal control of the CO2 chemical capture process has always been an important part in academic fields. Model predictive control (MPC) is a very effective control strategy used for such process, but the most intractable problem is the lack of accurate and effective model. In this work, Aspen Plus and Aspen Plus Dynamics are used to establish the process of monoethanolamine (MEA) absorption of CO2 related models based on subspace identification. The nonlinear distribution of the system under steady-state operation is analyzed. Dynamic tests were carried out to understand the dynamic characteristics of the system under variable operating conditions. Systematic subspace identification on open-loop experimental data was performed. We designed a model predictive controller based on the identified model combined with the state-space equation using Matlab/Simulink to analyze the changes of the system under two different disturbances. The simulation results show that the control performance of the MPC algorithm is significantly better than that of the traditional proportion integral differential (PID) system, with excellent setpoint tracking ability and robustness, which improve the stability and flexibility of the system.

  PublisherOA PDFDOI

• Perceived uncertainties of characterization in LCA: a survey

  The International Journal of Life Cycle Assessment · 2020-08-03 · 35 citations

  articleOpen access1st author
  PublisherOA PDFDOI

Frequent coauthors

• Yuxiong Huang

  Tsinghua–Berkeley Shenzhen Institute

  12 shared

• Arturo A. Keller

  University of California, Santa Barbara

  12 shared

• Sangwon Suh

  11 shared

• Lijuan Zhao

  State Key Laboratory of Pollution Control and Resource Reuse

  6 shared

• Jianqiang Gu

  Tianjin University

  6 shared

• Huiling Zhang

  Chinese Academy of Sciences

  6 shared

• Xiaolei Qu

  State Key Laboratory of Pollution Control and Resource Reuse

  6 shared

• Rong Ji

  6 shared

Education

• Ph.D., Computer Science

  UC San Diego

Awards & honors

• UCSB Regents' Junior Faculty Fellowship Award