Title

Approximate Bayesian Decision‐making with Complex Data: Analysis of Forensic Fingerprint Data

Presenter Information/ Coauthors Information

Cedric Neumann, South Dakota State UniversityFollow

Presentation Type

Event

Track

Tools

Abstract

Bayesian inference allow us to use information contained in a dataset to update a prior belief about some parameter of interest (e.g., a population mean) and make some inferences about the value of the parameter. The result allows us to quantify the uncertainty about the value of the parameter in a more logical and coherent way than traditional frequentist techniques. Unfortunately, standard Bayesian methods cannot be applied in all scenarios. This is the case for many scenarios that require unreasonably complex models to describe the data and where the corresponding likelihood function cannot be derived. A class of methods, called Approximate Bayesian Computation (ABC), allows for approximate Bayesian inference to be performed in these scenarios. ABC methods are simulation based and allow for coherent decision-making. ABC methods can be very useful to analyze the results of experiments from a wide range of disciplines (animal science, plant science, healthcare, finance) where the data may be unbalanced, high-dimensional, or encapsulate many different variable types. Two examples of the application of ABC forensic evidence will be provided.

Start Date

5-2-2019 11:00 AM

End Date

5-2-2019 12:00 PM

This document is currently not available here.

Share

Import Event to Google Calendar

COinS
 
Feb 5th, 11:00 AM Feb 5th, 12:00 PM

Approximate Bayesian Decision‐making with Complex Data: Analysis of Forensic Fingerprint Data

Pasque 255

Bayesian inference allow us to use information contained in a dataset to update a prior belief about some parameter of interest (e.g., a population mean) and make some inferences about the value of the parameter. The result allows us to quantify the uncertainty about the value of the parameter in a more logical and coherent way than traditional frequentist techniques. Unfortunately, standard Bayesian methods cannot be applied in all scenarios. This is the case for many scenarios that require unreasonably complex models to describe the data and where the corresponding likelihood function cannot be derived. A class of methods, called Approximate Bayesian Computation (ABC), allows for approximate Bayesian inference to be performed in these scenarios. ABC methods are simulation based and allow for coherent decision-making. ABC methods can be very useful to analyze the results of experiments from a wide range of disciplines (animal science, plant science, healthcare, finance) where the data may be unbalanced, high-dimensional, or encapsulate many different variable types. Two examples of the application of ABC forensic evidence will be provided.