Invariant Risk Minimization Games

Abstract

The standard risk minimization paradigm of machine learning is brittle when operating in environments whose test distributions are different from the training distribution due to spurious correlations. Training on data from many environments and finding invariant predictors reduces the effect of spurious features by concentrating models on features that have a causal relationship with the outcome. In this work, we pose such invariant risk minimization as finding the Nash equilibrium of an ensemble game among several environments. By doing so, we develop a simple training algorithm that uses best response dynamics and, in our experiments, yields similar or better empirical accuracy with much lower variance than the challenging bi-level optimization problem of Arjovsky et al. (2019). One key theoretical contribution is showing that the set of Nash equilibria for the proposed game are equivalent to the set of invariant predictors for any finite number of environments, even with nonlinear classifiers and transformations. As a result, our method also retains the generalization guarantees to a large set of environments shown in Arjovsky et al. (2019). The proposed algorithm adds to the collection of successful game-theoretic machine learning algorithms such as generative adversarial networks.

References in corpus (5)

• Agnostic Federated Learning
• Domain-Adversarial Neural Networks
• On Learning Invariant Representation for Domain Adaptation
• On Causal and Anticausal Learning
• Support and Invertibility in Domain-Invariant Representations

Cited by in corpus (21)

• Domain Generalization using Causal Matching
• Gradient Starvation: A Learning Proclivity in Neural Networks
• Towards a Theoretical Framework of Out-of-Distribution Generalization
• Unshuffling Data for Improved Generalization
• Nonlinear Invariant Risk Minimization: A Causal Approach
• An Empirical Study of Invariant Risk Minimization
• Out-of-distribution Prediction with Invariant Risk Minimization: The Limitation and An Effective Fix
• Empirical or Invariant Risk Minimization? A Sample Complexity Perspective
• Learning Causal Models Online
• Invariant Policy Optimization: Towards Stronger Generalization in Reinforcement Learning
• Variational Disentanglement for Domain Generalization
• Out-of-Distribution Generalization Analysis via Influence Function
• Understanding Continual Learning Settings with Data Distribution Drift Analysis
• Linear Regression Games: Convergence Guarantees to Approximate Out-of-Distribution Solutions
• Cross-Domain Sentiment Classification with In-Domain Contrastive Learning
• Meta-Learned Invariant Risk Minimization
• The Connection between Out-of-Distribution Generalization and Privacy of ML Models
• Embracing the Dark Knowledge: Domain Generalization Using Regularized Knowledge Distillation
• Optimization-based Causal Estimation from Heterogenous Environments
• Learning Diverse Representations for Fast Adaptation to Distribution Shift
• An Empirical Framework for Domain Generalization in Clinical Settings