Outside the Echo Chamber: Optimizing the Performative Risk

In performative prediction, predictions guide decision-making and hence can influence the distribution of future data. To date, work on performative prediction has focused on finding performatively stable models, which are the fixed points of repeated retraining. However, stable solutions can be far from optimal when evaluated in terms of the performative risk, the loss experienced by the decision maker when deploying a model. In this paper, we shift attention beyond performative stability and focus on optimizing the performative risk directly. We identify a natural set of properties of the loss function and model-induced distribution shift under which the performative risk is convex, a property which does not follow from convexity of the loss alone. Furthermore, we develop algorithms that leverage our structural assumptions to optimize the performative risk with better sample efficiency than generic methods for derivative-free convex optimization.

[1]  Aaron Roth,et al.  Strategic Classification from Revealed Preferences , 2017, EC.

[2]  O. Papaspiliopoulos.  High-Dimensional Probability: An Introduction with Applications in Data Science , 2020 .

[3]  Ilias Zadik,et al.  Orthogonal Machine Learning: Power and Limitations , 2017, ICML.

[4]  Brian Axelrod,et al.  Causal Strategic Linear Regression , 2020, ICML.

[5]  Yiling Chen,et al.  Learning Strategy-Aware Linear Classifiers , 2019, NeurIPS.

[6]  Roman Vershynin,et al.  High-Dimensional Probability , 2018 .

[7]  P. Robinson.  ROOT-N-CONSISTENT SEMIPARAMETRIC REGRESSION , 1988 .

[8]  Christian Hansen,et al.  Double/Debiased/Neyman Machine Learning of Treatment Effects , 2017, 1701.08687.

[9]  Navin Kartik,et al.  Improving Information from Manipulable Data , 2019, Journal of the European Economic Association.

[10]  Zhiwei Steven Wu,et al.  Causal Feature Discovery through Strategic Modification , 2020, ArXiv.

[11]  Nicole Immorlica,et al.  Maximizing Welfare with Incentive-Aware Evaluation Mechanisms , 2020, IJCAI.

[12]  Manuel Gomez-Rodriguez,et al.  Decisions, Counterfactual Explanations and Strategic Behavior , 2020, NeurIPS.

[13]  Nikolai Matni,et al.  A Tutorial on Concentration Bounds for System Identification , 2019, 2019 IEEE 58th Conference on Decision and Control (CDC).

[14]  Max Simchowitz,et al.  Naive Exploration is Optimal for Online LQR , 2020, ICML.

[15]  D. Drusvyatskiy,et al.  Stochastic Optimization with Decision-Dependent Distributions , 2020, Math. Oper. Res..

[16]  Kuang Xu,et al.  Experimenting in Equilibrium , 2019, Manag. Sci..

[17]  Celestine Mendler-Dünner,et al.  Performative Prediction , 2020, ICML.

[18]  Charles R. Johnson,et al.  Topics in Matrix Analysis , 1991 .

[19]  Adam Tauman Kalai,et al.  Online convex optimization in the bandit setting: gradient descent without a gradient , 2004, SODA '05.

[20]  Mark A. McComb.  Comparison Methods for Stochastic Models and Risks , 2003, Technometrics.

[21]  Zhiwei Steven Wu,et al.  Gaming Helps! Learning from Strategic Interactions in Natural Dynamics , 2021, AISTATS.

[22]  R. Z. Khasʹminskiĭ,et al.  Statistical estimation : asymptotic theory , 1981 .

[23]  Celestine Mendler-Dünner,et al.  Stochastic Optimization for Performative Prediction , 2020, NeurIPS.

[24]  Sheldon M. Ross,et al.  Stochastic Processes , 2018, Gauge Integral Structures for Stochastic Calculus and Quantum Electrodynamics.

[25]  Ohad Shamir,et al.  On the Complexity of Bandit and Derivative-Free Stochastic Convex Optimization , 2012, COLT.

[26]  W. Newey,et al.  Semiparametric Efficiency Bounds , 1990 .

[27]  Stochastic Orders , 2008 .

[28]  H. Theil.  A Note on Certainty Equivalence in Dynamic Planning , 1957 .

[29]  Ohad Shamir,et al.  Learnability, Stability and Uniform Convergence , 2010, J. Mach. Learn. Res..

[30]  Moritz Hardt,et al.  Strategic Classification is Causal Modeling in Disguise , 2019, ICML.

[31]  H. Simon,et al.  Dynamic Programming Under Uncertainty with a Quadratic Criterion Function , 1956 .

[32]  Nicole Immorlica,et al.  The Disparate Effects of Strategic Manipulation , 2018, FAT.

[33]  Benjamin Recht,et al.  Certainty Equivalence is Efficient for Linear Quadratic Control , 2019, NeurIPS.

[34]  Krikamol Muandet,et al.  Fair Decisions Despite Imperfect Predictions , 2019, AISTATS.

[35]  Bernhard Schölkopf,et al.  On the design of consequential ranking algorithms , 2020, UAI.

[36]  Christos H. Papadimitriou,et al.  Strategic Classification , 2015, ITCS.

[37]  J. Robins,et al.  Double/Debiased Machine Learning for Treatment and Structural Parameters , 2017 .

[38]  R. Hogg,et al.  On adaptive estimation , 1984 .

[39]  Ludger Rüschendorf,et al.  On the optimal stopping values induced by general dependence structures , 2001, Journal of Applied Probability.

[40]  Lin Xiao,et al.  Optimal Algorithms for Online Convex Optimization with Multi-Point Bandit Feedback. , 2010, COLT 2010.

[41]  Solomon Messing,et al.  Projecting Confidence: How the Probabilistic Horse Race Confuses and Demobilizes the Public , 2019, The Journal of Politics.

[42]  B. Levit,et al.  On the Efficiency of a Class of Non-Parametric Estimates , 1976 .

[43]  Anca D. Dragan,et al.  The Social Cost of Strategic Classification , 2018, FAT.

[44]  Lin F. Yang,et al.  Model-Based Reinforcement Learning with a Generative Model is Minimax Optimal , 2019, COLT 2020.

[45]  Evan Munro.  Learning Personalized Policy with Strategic Agents , 2020 .

[46]  A. Müller,et al.  Comparison Methods for Stochastic Models and Risks , 2002 .

[47]  Shlomi Hod,et al.  Performative Prediction in a Stateful World , 2020, ArXiv.

[48]  David C. Parkes,et al.  From Predictions to Decisions: Using Lookahead Regularization , 2020, NeurIPS.