|
Activity Number:
|
107
- The ABC of Making an Impact
|
|
Type:
|
Invited
|
|
Date/Time:
|
Monday, July 29, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
|
Abstract #300237
|
|
|
Title:
|
ABC and Forests: Where We Are and Where We Are Going
|
|
Author(s):
|
Louis Raynal* and Alice Cleynen and Jean-Michel Marin
|
|
Companies:
|
Alexander Grothendieck Montpellier Institute, University of Montpellier and Alexander Grothendieck Montpellier Institute, University of Montpellier and Alexander Grothendieck Montpellier Institute, University of Montpellier
|
|
Keywords:
|
Approximate Bayesian Computation;
random forests;
instance-based methods
|
|
Abstract:
|
Random forests (Breiman, 2001) provide good properties bypassing two major difficulties of most ABC methods: the preliminary selection of a vector of informative statistics summarizing raw data, and the calibration of a tolerance level separating acceptance from rejection of simulated parameter values. We present the ABC Random Forests approach (ABC-RF) in the regression setting and comparisons with earlier ABC techniques as well as adaptive sequential strategies. In addition to require very few calibration, ABC-RF offers a good trade-off in terms of quality of point estimator precision and credible interval estimations. Finally, in some application fields - as population genetics - only one observed data is of interest. Instance-based methods take into account the additional information provided by this test data to potentially improve the prediction accuracy. This is why we discuss whether or not a gain can be achieved by using instance-based versions of random forests instead of the usual one.
|
Authors who are presenting talks have a * after their name.
