Abstract:Invariant risk minimization (IRM) has received increasing attention as a way to acquire environment-agnostic data representations and predictions, and as a principled solution for preventing spurious correlations from being learned and for improving models' out-of-distribution generalization. Yet, recent works have found that the optimality of the originally-proposed IRM optimization (IRM) may be compromised in practice or could be impossible to achieve in some scenarios. Therefore, a series of advanced IRM algorithms have been developed that show practical improvement over IRM. In this work, we revisit these recent IRM advancements, and identify and resolve three practical limitations in IRM training and evaluation. First, we find that the effect of batch size during training has been chronically overlooked in previous studies, leaving room for further improvement. We propose small-batch training and highlight the improvements over a set of large-batch optimization techniques. Second, we find that improper selection of evaluation environments could give a false sense of invariance for IRM. To alleviate this effect, we leverage diversified test-time environments to precisely characterize the invariance of IRM when applied in practice. Third, we revisit (Ahuja et al. (2020))'s proposal to convert IRM into an ensemble game and identify a limitation when a single invariant predictor is desired instead of an ensemble of individual predictors. We propose a new IRM variant to address this limitation based on a novel viewpoint of ensemble IRM games as consensus-constrained bi-level optimization. Lastly, we conduct extensive experiments (covering 7 existing IRM variants and 7 datasets) to justify the practical significance of revisiting IRM training and evaluation in a principled manner.

Realistic Thoughts on IRM in Enterprises under Intelligence Meaning

Enterprise Information System and Information Rationalization

From "Information Service" to "Intelligence Intervention": Thinking on the Function of Intelligence from Decision Mistakes

Rethinking the construction of intelligent factory in process industry

Intellectual Management of Enterprise

Exploring the Role of Intellectual Capital in Supply Chain Intelligence Integration

Research on a Unified Approach for Enterprise Information Resource Management

What Is Missing in IRM Training and Evaluation? Challenges and Solutions

An Enterprise Model and the Organisation of ERP

On the logic and framework of enterprise management reform in the era of artificial intelligence

Research On Enterprise Informational Resources Allocation: Operational Echanism, Strategic Procedure And Countermeasures

Enterprise Modeling and the Elements of Reference Architecture Study

Research on intelligent operation and maintenance management method of enterprise IT

The concept and management of the risk of information technology

Innovative Use of Enterprise Information System:A Review and Future Directions

Information retrieval: a view from the Chinese IR community

Intelligence Failure——An Important Topic in Western Intelligence Studies and Its Inspirations to its Practices in China

Research on IRMS of Listed Companies

Research on Information Science Reconsidered

CIS Analysis Model:How to Transform Information to Intelligence

Enterprise informationization project evaluation framework study: requirement,problem and solution