Causal Modeling, University of Washington, Seattle

• Research project supervised by Dr. Thomas Richardson (Department of Statistics, University of Washington).
• Studied the causal effect of education on future earnings in the United States using compulsory schooling laws, with Quarter of Birth (QOB) as an instrumental variable.
• Implemented econometric and causal inference methods, including:
  • Two-Stage Least Squares (2SLS) to estimate the return associated with an additional year of education.
  • Fast Causal Inference (FCI) to assess instrumental-variable assumptions and investigate the possibility of hidden confounding.
  • Nonparametric polytope methods to obtain bounds on causal effects under weaker assumptions.
• Findings highlighted the contrast between precise parametric estimates and substantially wider nonparametric bounds, emphasizing the limitations of weak instruments and strong modeling assumptions.
• Poster presented at the HUB, University of Washington.
• Project Report | Poster

Math to Power Industry Workshop, Pacific Institute for the Mathematical Sciences (PIMS), in collaboration with Finite Carbon, USA

• Collaborated in a team of five to develop a machine learning pipeline for detecting forest change from satellite imagery in an industry-focused setting. Team members: Arman Jahangiri, Brian Andres Zambrano Luna, Issac Asamoah, Patrik Coulibaly, and Yasaman Shahhosseini.
• Worked with industry partner Finite Carbon under the mentorship of Bahareh Yekkehkhany.
• Acquired and processed ESA Sentinel-2 and Landsat imagery using Google Earth Engine and the United States Geological Survey.
• Performed data preprocessing steps including cloud masking, feature engineering, and geospatial alignment across satellite sources.
• Built and optimized a Random Forest classifier using grid search for model selection and performance improvement.
• Presented findings and actionable recommendations to Finite Carbon and other industry stakeholders during the program’s graduation event.
• Certificate

Contributed Talk, 2024 Ottawa Mathematics and Statistics Conference (OMSC), University of Ottawa

• Presented research on binary classification in high-dimensional settings with sparse and weak signals.
• Introduced classification fundamentals and motivating applications, including genomic data analysis and spam detection.
• Discussed key statistical classifiers, including logistic regression, k-nearest neighbors, Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA).
• Developed a theoretical framework addressing challenges arising from the curse of dimensionality.
• Presented impossibility regions and new theoretical results characterizing the optimality and limitations of QDA under specific signal regimes.
• Included computational analyses and alternative classification strategies supported by theoretical insights and performance comparisons.
• Slides (Contributed Talk) | Slides (Three-Minute Thesis Competition - Second Place Award)

Project in Nonparametric Statistics (Group Project), University of Calgary, Calgary

• Collaborated on a group project analyzing the Shapiro–Wilk test and selected extensions for assessing normality in data.
• Team members: Asal Rahmani Lahoot, Arman Jahangiri, Shiyu Sun, and Jessica Grant.
• Conducted a literature review of related work on normality testing.
• Studied the theoretical construction of the Shapiro–Wilk test statistic and reproduced benchmark results from the literature.
• Implemented the methods in R and compared the Shapiro–Wilk procedure with other normality tests in practice.
• Slides

Project in Generalized Linear Models (Group Project), University of Calgary, Calgary

• Conducted a data analysis project on a heart disease dataset with a focus on binomial regression models combined with feature-selection methods.
• Team members: Arman Jahangiri, Mojtaba Kanani Sarcheshmeh, and David Yang.
• Applied LASSO, forward selection, and backward selection to identify relevant predictors and improve model interpretability.
• Estimated heart disease likelihood and predicted chest pain type using patient-level covariates.
• Analyzed the relative risk associated with clinical variables and explored associations among heart-health factors.
• Performed preprocessing, exploratory data analysis, and visualization to support interpretation and communication of results.
• Project Report | Slides

Contributed Talk, 2023 Data Science Boot Camp, University of Saskatchewan

• Delivered a contributed talk on semiparametric varying-coefficient partially linear models (VC-PLMs) and their role in modeling complex data relationships.
• Presented theoretical ideas related to profile likelihood ratio tests, asymptotic properties, and bandwidth selection methods based on the work of Fan and Huang (2005).
• Emphasized the flexibility of VC-PLMs for robust semiparametric inference in applied statistical settings.
• Slides