Projects
Building Competent Models: Ensembles for Unlearning Single Feature Artifacts
Large NLI datasets often exhibit word-label imbalances, commonly referred to as single feature artifacts, which pose challenges for training robust language inference models. In this project, I apply the competency problems framework to identify these artifacts within the SNLI dataset. I analyze their impact on entailment predictions made by an ELECTRA-small model and present evidence that training ELECTRA-small in a mixed capacity ensemble with a unigram bag-of-words multi-layer perceptron model can effectively mitigate single feature bias.
Replication Package: Projecting the Impact of Rising Temperatures: The Role of Macroeconomic Dynamics.
This repository provides the replication package for the study 'Projecting the Impact of Rising Temperatures: The Role of Macroeconomic Dynamics,' published in the IMF Economic Review in 2023. Authored by Gregory Casey, Stephie Fried, and Ethan Goode, the study explores how temperature fluctuations affect economic growth, distinguishing between temporary impacts on output per capita and more enduring effects. The replication package includes the data, code, and documentation necessary to reproduce the study's key findings and projections.