论文信息 - Feedback Detection for Live Statistical Predictors

Feedback Detection for Live Statistical Predictors

A statistical predictor that is deployed in a live production system may perturb the features it uses to make predictions. Such a feedback loop can occur, for example, when a model that predicts a certain type of behavior ends up causing the behavior it predicts, thus creating a self-fulfilling prophecy. This paper analyzes statistical feedback detection as a causal inference problem, and proposes a local randomization scheme that can be used to detect feedback in real-world problems. We apply our method to a predictive model currently in use at an internet company.

Stefan Wager | Amir Najmi | Nick Chamandy | Omkar Muralidharan

[1] Jon Danielsson,et al. The emperor has no clothes: Limits to risk modelling , 2000 .

[2] P. J. Huber. The behavior of maximum likelihood estimates under nonstandard conditions , 1967 .

[3] J. Stock,et al. Instrumental Variables Regression with Weak Instruments , 1994 .

[4] G. Wahba. Spline models for observational data , 1990 .

[5] B. Silverman,et al. Nonparametric Regression and Generalized Linear Models: A roughness penalty approach , 1993 .

[6] P. Holland. Statistics and Causal Inference , 1985 .

[7] Tomaso A. Poggio,et al. Regularization Networks and Support Vector Machines , 2000, Adv. Comput. Math..

[8] Tomaso A. Poggio,et al. Regularization Theory and Neural Networks Architectures , 1995, Neural Computation.

[9] Fabrizio Ferraro,et al. Economics Language and Assumptions: How Theories Can Become Self-Fulfilling , 2003 .

[10] D. Rubin. Causal Inference Using Potential Outcomes , 2005 .