Learning Causal Models Online

Abstract

Predictive models -- learned from observational data not covering the complete data distribution -- can rely on spurious correlations in the data for making predictions. These correlations make the models brittle and hinder generalization. One solution for achieving strong generalization is to incorporate causal structures in the models; such structures constrain learning by ignoring correlations that contradict them. However, learning these structures is a hard problem in itself. Moreover, it's not clear how to incorporate the machinery of causality with online continual learning. In this work, we take an indirect approach to discovering causal models. Instead of searching for the true causal model directly, we propose an online algorithm that continually detects and removes spurious features. Our algorithm works on the idea that the correlation of a spurious feature with a target is not constant over-time. As a result, the weight associated with that feature is constantly changing. We show that by continually removing such features, our method converges to solutions that have strong generalization. Moreover, our method combined with random search can also discover non-spurious features from raw sensory data. Finally, our work highlights that the information present in the temporal structure of the problem -- destroyed by shuffling the data -- is essential for detecting spurious features online.

Spurious features, causal models, online learning, random search, non-iid

References in corpus (9)

• Adam: A Method for Stochastic Optimization
• Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
• Invariant Risk Minimization
• Out-of-Distribution Generalization via Risk Extrapolation (REx)
• Random Search and Reproducibility for Neural Architecture Search
• A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms
• Learning Neural Causal Models from Unknown Interventions
• Invariant Risk Minimization Games
• From Dependence to Causation

Cited by in corpus (6)

• General Value Function Networks
• A Theory of Label Propagation for Subpopulation Shift
• Resolving Spurious Correlations in Causal Models of Environments via Interventions
• On Invariance Penalties for Risk Minimization
• Causality in Neural Networks -- An Extended Abstract
• Individual and Collective Autonomous Development