# Finding emergence in data: causal emergence inspired dynamics learning

**Authors:** Mingzhe Yang, Zhipeng Wang, Kaiwei Liu, Yingqi Rong, Bing Yuan, Jiang Zhang  
**arXiv:** [2308.09952](https://arxiv.org/abs/2308.09952)  
**Date:** 2023-08-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2308.09952v1/

## Key points

- The paper addresses the challenge of modeling complex dynamical systems in a data-driven manner
- It proposes a machine learning framework inspired by causal emergence theory
- The framework captures emergent dynamics at the macro-level and quantifies emergence based on available data
- Experimental results show that it successfully captures emergent patterns and learns coarse-graining strategies
- It quantifies the degree of causal emergence present in the data, providing insights into how emergent behaviors arise from micro-level interactions
- The framework demonstrates superior generalization ability to new scenarios beyond training data

## Summary

The paper titled "Finding emergence in data: causal emergence inspired dynamics learning" addresses the challenge of modeling complex dynamical systems in a data-driven manner. These systems often exhibit emergent behaviors and properties that cannot be directly captured by micro-level observational data. To overcome this limitation, the authors propose a machine learning framework that can effectively capture emergent dynamics at the macro-level and quantify emergence based on the available data. The framework draws inspiration from the theory of causal emergence and aims to learn macro-dynamics within an emergent latent space. It achieves this by maximizing the effective information (EI) to obtain a macro-dynamics model with stronger causal effects. The authors demonstrate the effectiveness of their proposed framework through experimental results on both simulated and real data. Notably, the framework successfully captures emergent patterns in the data and also learns the coarse-graining strategy, which is crucial for understanding complex systems. Additionally, it quantifies the degree of causal emergence present in the data, providing insights into how emergent behaviors arise from underlying micro-level interactions. Furthermore, experiments conducted on environments different from the training dataset highlight its superior generalization ability to new scenarios beyond those used for training. Overall, this paper presents a novel approach to modeling complex dynamical systems by leveraging machine learning techniques and drawing inspiration from causal emergence theory. The results demonstrate its effectiveness in capturing emergent patterns, learning coarse-graining strategies, quantifying causal emergence, and generalizing well to new environments.