iCubed Seminar: Brian Barr (Capital One) & Matt Harrington (Columbia SIPA)
Thursday, December 2, 2021
3:00 pm - 4:00 pm
Brian Barr, Machine Learning Researcher, Capital One
Matthew Harrington, PhD Student, School of International and Public Affairs (SIPA), Columbia University
Moderated By: John Hyde, DSI Assistant Director of Career Development and Alumni Services
Thursday, December 2 (3:00 PM – 4:00 PM ET) – Virtual
Counterfactual explanations represent the minimal change to a data sample that alters its predicted classification, typically from an unfavorable initial class to a desired target class. Counterfactuals can help answer questions like “what needs to change for this loan application to get accepted?”. A number of recently proposed approaches to counterfactual generation are computationally intensive and provide unconvincing explanations.
We will discuss a new method dubbed SharpShooter, that starts by creating a projection of the input that classifies as the target class. Counterfactual candidates are then generated in latent space on the interpolation line between the input and its projection. We demonstrate that our framework translates core characteristics of a sample to its counterfactual through the use of learned representations. In addition, we show that SharpShooter is competitive across common quality metrics, excels at measures of realism, while being three orders of magnitude faster than comparable methods – making it well-suited for high velocity machine learning applications which require timely explanations.