Скачать или смотреть ISBA WM 2021: A. Carriquiry- CSI Bayes: Comparing and interpreting images in forensic evaluations

Foundational Lecture
Speaker
Alicia Carriquiry, Iowa State University, Distinguished Professor and President's Chair in Statistics
Chair
Sylvia Fruhwirth-Schnatter, WU Vienna University of Economics and Business, Full Professor of Applied Statistics and Econometrics

Description
CSI Bayes: Comparing and interpreting images in forensic evaluations

In 1977, Lindley proposed using Bayes Factors (BF) to address the question of source: do two items, one found at a crime scene, and another recovered from the suspect, have a common origin? The BF approach, and its close relative, the Likelihood Ratio (LR) approach, have found their way into forensic practice, at least when the evidence can be summarized using numbers. This is the case with forensic DNA analysis or with the analysis of evidence such as glass and paint fragments, where we can get measurements of trace amounts of various components. In this talk, I discuss the case of pattern comparison evidence such as footwear treads, handwriting or expended ammunition, where the "data" come in the form of an image. While in some cases we can obtain numerical measurements from the images and implement Lindley’s approach, oftentimes we quantify the similarity between two items using algorithms such as a neural net or a random forest. We can still build what is known as a score-based LR, but there is a price to pay.