刑事证据评估的统计学习算法（LEARNING ALGORITHMS TO EVALUATE FORENSIC EVIDENCE）
报告人： Alicia Carriquiry, Iowa State University
时间：2018-05-21 16:00 ~ 17:00
The evaluation of evidence arises in criminal and civil judicial proceedings. In recent years, the scientific validity of most forensic analyses has been questioned, and there is now a push to develop the scientific and statistical underpinnings of various type of forensic tools. Pattern evidence including fingerprints, shoe out-sole impressions, striations on bullets and breech faces and others is particularly challenging, because this type of evidence is typically represented as an image, and does not lend itself to the traditional statistical modeling approach. Even numerical evidence such as the chemical composition of glass or paint is difficult to model, because it is often highly multi-dimensional.
In this talk, we focus on the question of source: do two evidence items have a common source? To answer this question, we develop non-parametric classification algorithms and show their application on glass and shoe out-sole impressions. We argue that learning algorithms enable forensic practitioners to quantify the degree of similarity between two items of evidence. Furthermore, by computing a score-based likelihood ratio, practitioners can assess the weight of the evidence in support of the prosecution’s or the defense’s propositions.
About the Speaker: