Corpus ID: 211258809

Learning From Strategic Agents: Accuracy, Improvement, and Causality

@article{Shavit2020LearningFS,
  title={Learning From Strategic Agents: Accuracy, Improvement, and Causality},
  author={Yonadav Shavit and Benjamin L. Edelman and Brian Axelrod},
  journal={ArXiv},
  year={2020},
  volume={abs/2002.10066}
}
In many predictive decision-making scenarios, such as credit scoring and academic testing, a decision-maker must construct a model (predicting some outcome) that accounts for agents' incentives to "game" their features in order to receive better decisions. Whereas the strategic classification literature generally assumes that agents' outcomes are not causally dependent on their features (and thus strategic behavior is a form of lying), we join concurrent work in modeling agents' outcomes as a… Expand
Gaming Helps! Learning from Strategic Interactions in Natural Dynamics
TLDR
It is shown that even simple behavior on the learner’s part allows her to simultaneously accurately recover the meaningful features and incentivize agents to invest in these meaningful features, providing incentives for improvement. Expand
Strategic Classification is Causal Modeling in Disguise
TLDR
This work develops a causal framework for strategic adaptation and proves any procedure for designing classifiers that incentivize improvement must inevitably solve a non-trivial causal inference problem. Expand
Performative Prediction
TLDR
This work develops a risk minimization framework for performative prediction bringing together concepts from statistics, game theory, and causality, and gives the first sufficient conditions for retraining to overcome strategic feedback effects. Expand
Classification with Few Tests through Self-Selection
TLDR
It is shown that as long as the good and bad agents have different distributions, the principal can always achieve perfect accuracy, meaning good agents are accepted with probability 1, and bad ones are rejected with probability1. Expand
From Predictions to Decisions: Using Lookahead Regularization
TLDR
Look-ahead regularization is introduced which, by anticipating user actions, encourages predictive models to also induce actions that improve outcomes, and carefully tailors the uncertainty estimates governing confidence in this improvement to the distribution of model-induced actions. Expand
How to Learn when Data Reacts to Your Model: Performative Gradient Descent
TLDR
Under regularity assumptions on the performative loss, PerfGD is the first algorithm which provably converges to an optimal point and is simple to use. Expand

References

SHOWING 1-10 OF 17 REFERENCES
The disparate equilibria of algorithmic decision making when individuals invest rationally
TLDR
It is shown that decoupling achieves optimal outcomes in the realizable case but has discrepant effects that may depend on the initial conditions otherwise and subsidizing the cost of investment is shown to create better equilibria for the disadvantaged group even in the absence of realizability. Expand
Strategic Adaptation to Classifiers: A Causal Perspective
TLDR
This work frames the problem of adaptation as performing interventions in a causal graph, and articulate a formal distinction between gaming and improvement, and formalizes strategic classification in a new way that recognizes that the individual may improve, rather than only game. Expand
Optimal Decision Making Under Strategic Behavior
TLDR
This paper uses the theory of optimal transport to characterize how strategic investment of effort by individuals leads to a change in the feature distribution at a population level and derives an efficient greedy algorithm that is guaranteed to find locally optimal decision policies in polynomial time. Expand
Fairness is not static: deeper understanding of long term fairness via simulation studies
TLDR
An extensible open-source software framework for implementing fairness-focused simulation studies and further reproducible research is provided, and it is shown that static or single-step analyses do not give a complete picture of the long-term consequences of an ML-based decision system. Expand
Actionable Recourse in Linear Classification
TLDR
An integer programming toolkit is presented to measure the feasibility and difficulty of recourse in a target population, and generate a list of actionable changes for a person to obtain a desired outcome, and illustrate how recourse can be significantly affected by common modeling practices. Expand
How Do Classifiers Induce Agents to Invest Effort Strategically?
TLDR
A model for how strategic agents can invest effort in order to change the outcomes they receive is developed, and a tight characterization of when such agents can be incentivized to invest specified forms of effort into improving their outcomes is given. Expand
Strategic Classification
TLDR
This paper formalizes the problem, and pursue algorithms for learning classifiers that are robust to gaming, and obtains computationally efficient learning algorithms which are near optimal, achieving a classification error that is arbitrarily close to the theoretical minimum. Expand
Strategyproof Linear Regression in High Dimensions
TLDR
This paper focuses on the ubiquitous problem of linear regression, where strategyproof mechanisms have previously been identified in two dimensions, and finds a family of group strategyproof linear regression mechanisms in any number of dimensions, which are called generalized resistant hyperplane mechanisms. Expand
Causal discovery with continuous additive noise models
TLDR
If the observational distribution follows a structural equation model with an additive noise structure, the directed acyclic graph becomes identifiable from the distribution under mild conditions, which constitutes an interesting alternative to traditional methods that assume faithfulness and identify only the Markov equivalence class of the graph, thus leaving some edges undirected. Expand
Contextual Bandits with Linear Payoff Functions
TLDR
An O (√ Td ln (KT ln(T )/δ) ) regret bound is proved that holds with probability 1− δ for the simplest known upper confidence bound algorithm for this problem. Expand
...
1
2
...