# Predictively Encoded Graph Convolutional Network for Noise-Robust Skeleton-based Action Recognition

**Authors:** Jongmin Yu, Yongsang Yoon, Moongu Jeon  
**arXiv:** [2003.07514](https://arxiv.org/abs/2003.07514)  
**Date:** 2020-03-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2003.07514v1/

## Key points

- Graph convolutional networks (GCNs) have shown impressive performance in skeleton-based action recognition
- The proposed method addresses the issue of incomplete or noisy skeletons in real scenarios
- The approach maximizes mutual information between normal and noisy skeletons using predictive coding
- Normal and noise skeleton features are mapped into compact distributed vector representations through non-linear mapping functions
- The model is trained to preserve the mutual information between these representations
- Experimental results demonstrate superior performance compared to existing methods, particularly in handling noise and accurately modeling density ratios
- The method is compared with VA-LSTM and Clips+CN approaches, highlighting its advantages

## Summary

In skeleton-based action recognition, graph convolutional networks (GCNs) have shown impressive performance by modeling human body skeletons using nodes and connections. To address the issue of incomplete or noisy skeletons which may not be provided in real scenarios, the authors propose a robust skeleton-based action recognition method that can handle noise information in skeleton features. The key insight of their approach is to maximize the mutual information between normal and noisy skeletons using a predictive coding manner. They map normal and noise skeleton features into compact distributed vector representations through non-linear mapping functions and train the model to preserve the mutual information between these representations. The experimental results demonstrate its superior performance compared to existing methods when dealing with incomplete or noisy skeletons. They compare their method with VA-LSTM and Clips+CN approaches, highlighting its advantages such as better performance in handling noise and its ability to model density ratios accurately. Overall, this work presents a novel approach for robust skeleton-based action recognition that can effectively handle noise information in skeleton features.