HANNAH LU

ABOUT

DATA, COMPUTING, & POLICY

Hello! I currently work as a Fellow at the Science and Technology Policy Institute in Washington, DC. I graduated from Harvey Mudd College in 2024 with a joint major in Computer Science & Mathematics, as well as a concentration in Environment, Economics, & Politics.

My passion lies in environmental and social policy, though my background is perhaps a bit unconventional; prior to my transition to policy, I spent many years working with machine learning, data analysis, modelling, and computer programming.

This portfolio contains a select record of my past work. Many involve machine learning, and most are oriented towards social impact.


TECHNICAL SKILLS

PROGRAMMING

  • Python
    • Tensorflow
    • PyTorch
    • jax
    • scikit-learn
    • OpenCV
    • NumPy
    • pandas
  • R
  • C++
  • JAVA
  • Haskell

COURSEWORK

  • Policy Lab
  • Public Policy Analysis
  • Environmental Economics
  • Comparative Environmental Policy
  • Transportation Technology & Climate Policy
  • Mathematics of Machine Learning
  • Information-Theoretic Foundations of Machine Learning
  • Computer Vision
  • Algorithms
  • Data Structures & Program Development
  • Reinforcement Learning
  • Real Analysis I
  • Abstract Algebra I
  • Scientific Computing
  • Linear Algebra
  • Probability & Statistics
  • Differential Equations
  • Single & Multivariable Calculus
  • Discrete Mathematics

PORTFOLIO

Nature-Based Solutions: Evidence for Hazard Risk Reduction and Ecosystem Services
FALL 2024

STPI report on nature-based solutions, sponsored by the White House Office of Science and Technology Policy.

Contributions include writing, editing, and fact-checking.

Automated data extraction for debt collection policy analysis
SPRING 2024

Side project for the Claremont McKenna Policy Lab capstone project on debt collection court cases.

Through this team-based policy analysis project, we investigated how defendant demographics impacted the outcomes of debt collection court cases in California. The project was conducted in partnership with the Los Angeles County Superior Court and the RAND Institute for Civil Justice.

In order to accelerate data collection, I developed a program to automatically extract data from scanned legal documents, which was able to process 350 cases an hour. The pipeline leveraged computer vision techniques and optical character recognition, eliminating the need for manual data collection, which originally took hundreds of people-hours.

Other contributions include literature reviews, data analysis and visualization, and writing.

Explainability and feature bias in video classifiers
SUMMER 2023 – FALL 2023

NSF REU-funded research project under Professor Calden Wloka.

Investigated biases towards static and dynamic features in machine-learning models trained for video classification.

As an undergraduate researcher, I implemented key functionality in the analysis pipeline to visualize the behavior of arbitrary neural network video classifiers, ran experiments to measure feature biases, and probed a key mathematical error that I uncovered across many published code implementations of a popular model visualization tool. Additionally, I strategized research goals and coordinated fellow team members to achieve priorities in parallel.

GNN architecture
Forecasting atmospheric dynamics with graph neural networks
FALL 2022 – FALL 2023

Research project under Professor Sarah Kavassalis.

Investigating the impact of explicitly encoding spatial relationships in graph neural networks to forecast the evolution of nonlinear atmospheric dynamics from the Lorenz-96 model.

FLASH ASL spelled in the ASL alphabet
Flash-ASL
WINTER 2022 – WINTER 2023

Digital flashcards tailored for practicing American Sign Language (ASL).

Developed using Flask and hosted on AWS.

Folium city map
Predicting travel characteristics
SUMMER 2022

Software SULI internship at the National Renewable Energy Laboratory under K. Shankari.

As a summer intern, I implemented and analyzed candidate machine learning pipelines, using clustering and random forests, to predict characteristics of human travel such as mode of transportation and purpose of travel. To this end, I also developed a two-step supervised clustering pipeline which uses DBSCAN and SVMs to cluster trips geospatially.

These predictions can make unlabeled research data usable for aggregate analysis; they can also be used as label suggestions in the NREL OpenPATH data-collection tool to streamline the labeling process and reduce user burden.

social network graph
Network analysis of fake news propagation
SPRING 2022

Research project under Professor Susan Martonosi.

Developed simulations to analyze the propagation of information in a social network; examined how the spread of news was affected by the truthfulness and political bias of news articles, as well as readership bias and social network structure.

neural networks thumbnail
Neural networks
FALL 2020, FALL 2021

Image classification of animal species using convolutional neural networks, data augmentation, and transfer learning.

pollution thumbnail
Air pollution prediction via machine learning
FALL 2021

Exploration of classical machine learning algorithms for HMC's CS189 Programming Practicum elective.

Project goal: understand how classical machine learning algorithms work, and apply them to predict air pollution levels on the ground, based on weather data and satellite observations of pollutants. Dataset from Zindi.

PO sustainability report thumbnail
Pomona College Sustainability Report
SUMMER 2021

Data analysis and design at Pomona's Sustainability Office.

This is the annual evaluation of Pomona's performance on sustainability metrics, newly presented as an interactive report on Tableau. The 2021 report provides a unique look into the College's 'baseline' measurements during a year of fully remote learning, and lends insight on the effects of human behavior versus infrastructure.

p-sync thumbnail
P-SYNC
SPRING 2021

Exploration of the capabilities of reservoir computing. Applications to forecasting the growth of digital communities.

CONTACT