Probabilistic elicitation of expert knowledge through assessment of computer simulations

We present a new method for probabilistic elicitation of expert knowledge using binary responses of human experts assessing simulated data from a statistical model, where the parameters are subject to uncertainty. The binary responses describe either the absolute realism of individual simulations or the relative realism of a pair of simulations in the two alternative versions of out approach. Each version provides a nonparametric representation of the expert belief distribution over the values of a model parameter, without demanding the assertion of any opinion on the parameter values themselves. Our framework also integrates the use of active learning to efficiently query the experts, with the possibility to additionally provide a useful misspecification diagnostic. We validate both methods on an automatic expert judging a binomial distribution, and on human experts judging the distribution of voters across political parties in the United States and Norway. Both methods provide flexible and meaningful representations of the human experts' beliefs, correctly identifying the higher dispersion of voters between parties in Norway.

[1]  Samuel Kaski,et al.  Interactive Elicitation of Knowledge on Feature Relevance Improves Predictions in Small Data Sets , 2016, IUI.

[2]  Neil D. Lawrence,et al.  Preferential Bayesian Optimization , 2017, ICML.

[3]  J. Berger The case for objective Bayesian analysis , 2006 .

[4]  Ian H. Witten,et al.  Interactive machine learning: letting users build classifiers , 2002, Int. J. Hum. Comput. Stud..

[5]  Anthony O'Hagan,et al.  Uncertainty in prior elicitations: a nonparametric approach , 2007 .

[6]  Michael Goldstein,et al.  Subjective Bayesian Analysis: Principles and Practice , 2006 .

[7]  Andreas Holzinger,et al.  Reasoning Under Uncertainty: Towards Collaborative Interactive Machine Learning , 2016, Machine Learning for Health Informatics.

[8]  Yoshua Bengio,et al.  Generative Adversarial Nets , 2014, NIPS.

[9]  Samuel Kaski,et al.  Interactive intent modeling , 2014, Commun. ACM.

[10]  Judith Rousseau,et al.  Combining expert opinions in prior elicitation , 2010 .

[11]  Takafumi Kanamori,et al.  Density Ratio Estimation in Machine Learning , 2012 .

[12]  Samuel Kaski,et al.  Interactive Prior Elicitation of Feature Similarities for Small Sample Size Prediction , 2016, UMAP.

[13]  Gilles Louppe,et al.  Approximating Likelihood Ratios with Calibrated Discriminative Classifiers , 2015, 1506.02169.

[14]  Ritabrata Dutta,et al.  Likelihood-free inference via classification , 2014, Stat. Comput..

[15]  Samuel Kaski,et al.  Knowledge elicitation via sequential probabilistic inference for high-dimensional prediction , 2016, Machine Learning.

[16]  John Paul Gosling,et al.  SHELF: The Sheffield Elicitation Framework , 2018 .

[17]  Desney S. Tan,et al.  Effective End-User Interaction with Machine Learning , 2011, AAAI.

[18]  Gilles Louppe,et al.  Likelihood-free MCMC with Amortized Approximate Ratio Estimators , 2019, ICML.

[19]  Scott Sanner,et al.  Gaussian Process Preference Elicitation , 2010, NIPS.

[20]  Sindhu R Johnson,et al.  Methods to elicit beliefs for Bayesian priors: a systematic review. , 2010, Journal of clinical epidemiology.

[21]  Jerry Alan Fails,et al.  Interactive machine learning , 2003, IUI '03.

[22]  Gilles Louppe,et al.  Likelihood-free MCMC with Approximate Likelihood Ratios , 2019, ArXiv.

[23]  Jukka Corander,et al.  Diagnosing model misspecification and performing generalized Bayes' updates via probabilistic classifiers , 2019, 1912.05810.

[24]  Jeremy E. Oakley,et al.  Uncertain Judgements: Eliciting Experts' Probabilities , 2006 .

[25]  Masashi Sugiyama,et al.  Density-ratio matching under the Bregman divergence: a unified framework of density-ratio estimation , 2012 .

[26]  Jukka Corander,et al.  Likelihood-Free Inference by Ratio Estimation , 2016, Bayesian Analysis.