Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Causal Invariance and Counterfactual Learning Driven Cooperative Game for Multi-Label Classification

Yijia Fan, Jusheng Zhang, Kaitong Cai, Jing Yang, Keze Wang

TL;DR

The paper tackles the difficulty of rare-label prediction and robustness to distribution shifts in multi-label classification by introducing the Causal Cooperative Game (CCG) framework. CCG models MLC as a multi-player cooperative game in which players learn true label dependencies via Neural Structural Equation Models, guided by a counterfactual curiosity reward and a causal invariance loss, with targeted rare-label enhancement. Extensive experiments on four text benchmarks show consistent improvements in Rare-F1 and robustness to distribution shifts, supported by ablation studies and qualitative analyses of learned causal graphs that demonstrate interpretability. The work highlights the value of integrating causal inference with cooperative game theory to build more robust, generalizable, and explainable MLC systems.

Abstract

Multi-label classification (MLC) remains vulnerable to label imbalance, spurious correlations, and distribution shifts, challenges that are particularly detrimental to rare label prediction. To address these limitations, we introduce the Causal Cooperative Game (CCG) framework, which conceptualizes MLC as a cooperative multi-player interaction. CCG unifies explicit causal discovery via Neural Structural Equation Models with a counterfactual curiosity reward to drive robust feature learning. Furthermore, it incorporates a causal invariance loss to ensure generalization across diverse environments, complemented by a specialized enhancement strategy for rare labels. Extensive benchmarking demonstrates that CCG substantially outperforms strong baselines in both rare label prediction and overall robustness. Through rigorous ablation studies and qualitative analysis, we validate the efficacy and interpretability of our components, underscoring the potential of synergizing causal inference with cooperative game theory for advancing multi-label learning.

Causal Invariance and Counterfactual Learning Driven Cooperative Game for Multi-Label Classification

TL;DR

The paper tackles the difficulty of rare-label prediction and robustness to distribution shifts in multi-label classification by introducing the Causal Cooperative Game (CCG) framework. CCG models MLC as a multi-player cooperative game in which players learn true label dependencies via Neural Structural Equation Models, guided by a counterfactual curiosity reward and a causal invariance loss, with targeted rare-label enhancement. Extensive experiments on four text benchmarks show consistent improvements in Rare-F1 and robustness to distribution shifts, supported by ablation studies and qualitative analyses of learned causal graphs that demonstrate interpretability. The work highlights the value of integrating causal inference with cooperative game theory to build more robust, generalizable, and explainable MLC systems.

Abstract

Multi-label classification (MLC) remains vulnerable to label imbalance, spurious correlations, and distribution shifts, challenges that are particularly detrimental to rare label prediction. To address these limitations, we introduce the Causal Cooperative Game (CCG) framework, which conceptualizes MLC as a cooperative multi-player interaction. CCG unifies explicit causal discovery via Neural Structural Equation Models with a counterfactual curiosity reward to drive robust feature learning. Furthermore, it incorporates a causal invariance loss to ensure generalization across diverse environments, complemented by a specialized enhancement strategy for rare labels. Extensive benchmarking demonstrates that CCG substantially outperforms strong baselines in both rare label prediction and overall robustness. Through rigorous ablation studies and qualitative analysis, we validate the efficacy and interpretability of our components, underscoring the potential of synergizing causal inference with cooperative game theory for advancing multi-label learning.

Paper Structure

This paper contains 38 sections, 4 equations, 3 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-Label Classification (MLC)
  4. The Rare Label Problem
  5. Causal Machine Learning
  6. Method
  7. Causality-Driven Multi-Label Cooperative Game Framework
  8. Causal Graph Construction and Prior Constraints
  9. Prior Constraints
  10. Causal Graph Learning Objective
  11. Player Decomposition and Causal Constraints
  12. Counterfactual Curiosity Reward Mechanism
  13. Counterfactual Consistency Reward
  14. Causal Invariance Loss Function
  15. Weighted Cross-Entropy Loss for Multi-Label Classification
  16. ...and 23 more sections

Figures (3)

  • Figure 1: A subgraph showing the learned (hypothesized) causal relationships between concepts related to cardiovascular disease. The nodes represent specific medical labels, and the edges and their accompanying red weights indicate the mutual influence and learned strength between these concepts.
  • Figure 2: The curve of mAP score and F1-score of CCG on 20 Newsgroups with the change of Number of Players
  • Figure 3: Performance comparison under simulated temporal distribution shift on the Reuters Corpus Volume 1 (RCV1) dataset. ID denotes In-Distribution test set (earlier period), and OOD denotes Out-of-Distribution test set (later period). $\Delta$ indicates the absolute performance drop from ID to OOD. Best OOD performance and smallest degradation are highlighted in bold.