Invariant Risk Minimization Games

By Kartik Ahuja, Karthikeyan Shanmugam, Kush R. Varshney, Amit Dhurandhar. 2020

299 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Learning explanations that are hard to vary (2020, background): “This motivates our notion of invariant mechanisms, and inspired related notions which have been proposed for robust regression (Rojas-Carulla et al., 2018; Heinze-Deml et al., 2018; Arjovsky et al., 2019; Hermann and Lampinen, 2020; Ahuja et al., 2020; Krueger et al., 2020).”
In Search of Lost Domain Generalization (2020, background): “Ahuja et al. [2020] analyze IRM under a game-theoretic perspective to develop an alternative algorithm.”
Learning to Balance Specificity and Invariance for In and Out of Domain Generalization (2020, background): “Recently, a novel variant of empirical risk minimization framework, called Invariant Risk Minimization (IRM) has been proposed in [2,1] to make machine learning models invariant to spurious correlations in data when training across multiple sources.”
Understanding Continual Learning Settings with Data Distribution Drift Analysis (2021, background): “For example, the ﬁeld of out-of-distribution (OOD) generalization aims at developing learning algorithms able to perform well on test distributions that are different from training distribution (Arjovsky et al., 2019; Ahuja et al., 2020; Krueger et al., 2021).”
Generalizing to Unseen Domains: A Survey on Domain Generalization (2021, background): “There are some further theoretical analyses on the success [112] and failure cases of IRM [111], and IRM has been extended to other tasks including text classification [110] and reinforcement learning [197].”
Domain Generalization using Causal Matching (2020, background): “…meta-learning (Li et al., 2018a; Balaji et al., 2018), dataset augmentation (Volpi et al., 2018; Shankar et al., 2018), parameter decomposition (Piratla et al., 2020; Li et al., 2017), and enforcing domain-invariance of the optimal P ( Y | Φ( x )) (Arjovsky et al., 2019; Ahuja et al., 2020).”
Gradient Starvation: A Learning Proclivity in Neural Networks (2020, background): “Recent work aims at identifying the reasons behind this phenomenon [97, 70, 22, 73, 51, 76, 100, 92, 83, 105, 42, 79, 4], while complementary work quantifies resulting shortcomings, including poor generalization to out-of-distribution (OOD) test data, reliance upon spurious correlations, and lack of”
Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development (2021, methodology): “We include the following domain generalization algorithms: MMD (Maximum Mean Discrepancy) [83] optimizes the similarities of maximum mean discrepancy across domains, CORAL (Correlation Alignment) [137] matches the mean and the covariance of features across domains; IRM (Invariant Risk Minimization) ”