Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AutoGCN -- Towards Generic Human Activity Recognition with Neural Architecture Search

Felix Tempel, Inga Strümke, Espen Alexander F. Ihlen

TL;DR

AutoGCN tackles skeleton-based HAR by introducing a generic neural architecture search framework that jointly optimizes GCN architectures and hyperparameters. It combines a domain-informed, diverse search space with a reinforcement controller and a knowledge reservoir to balance exploration and exploitation, yielding architectures that generalize across datasets. Empirically, AutoGCN outperforms random search and prior HAR NAS methods, achieving 95.5% Top-1 on NTU RGB+D 60 X-View and competitive results on NTU RGB+D 120, while using efficient training and without ensemble methods. The work demonstrates the importance of expressive input representations and a broad search space for robust HAR models, and points to future improvements via one-shot NAS and performance predictors to further accelerate the search.

Abstract

This paper introduces AutoGCN, a generic Neural Architecture Search (NAS) algorithm for Human Activity Recognition (HAR) using Graph Convolution Networks (GCNs). HAR has gained attention due to advances in deep learning, increased data availability, and enhanced computational capabilities. At the same time, GCNs have shown promising results in modeling relationships between body key points in a skeletal graph. While domain experts often craft dataset-specific GCN-based methods, their applicability beyond this specific context is severely limited. AutoGCN seeks to address this limitation by simultaneously searching for the ideal hyperparameters and architecture combination within a versatile search space using a reinforcement controller while balancing optimal exploration and exploitation behavior with a knowledge reservoir during the search process. We conduct extensive experiments on two large-scale datasets focused on skeleton-based action recognition to assess the proposed algorithm's performance. Our experimental results underscore the effectiveness of AutoGCN in constructing optimal GCN architectures for HAR, outperforming conventional NAS and GCN methods, as well as random search. These findings highlight the significance of a diverse search space and an expressive input representation to enhance the network performance and generalizability.

AutoGCN -- Towards Generic Human Activity Recognition with Neural Architecture Search

TL;DR

AutoGCN tackles skeleton-based HAR by introducing a generic neural architecture search framework that jointly optimizes GCN architectures and hyperparameters. It combines a domain-informed, diverse search space with a reinforcement controller and a knowledge reservoir to balance exploration and exploitation, yielding architectures that generalize across datasets. Empirically, AutoGCN outperforms random search and prior HAR NAS methods, achieving 95.5% Top-1 on NTU RGB+D 60 X-View and competitive results on NTU RGB+D 120, while using efficient training and without ensemble methods. The work demonstrates the importance of expressive input representations and a broad search space for robust HAR models, and points to future improvements via one-shot NAS and performance predictors to further accelerate the search.

Abstract

This paper introduces AutoGCN, a generic Neural Architecture Search (NAS) algorithm for Human Activity Recognition (HAR) using Graph Convolution Networks (GCNs). HAR has gained attention due to advances in deep learning, increased data availability, and enhanced computational capabilities. At the same time, GCNs have shown promising results in modeling relationships between body key points in a skeletal graph. While domain experts often craft dataset-specific GCN-based methods, their applicability beyond this specific context is severely limited. AutoGCN seeks to address this limitation by simultaneously searching for the ideal hyperparameters and architecture combination within a versatile search space using a reinforcement controller while balancing optimal exploration and exploitation behavior with a knowledge reservoir during the search process. We conduct extensive experiments on two large-scale datasets focused on skeleton-based action recognition to assess the proposed algorithm's performance. Our experimental results underscore the effectiveness of AutoGCN in constructing optimal GCN architectures for HAR, outperforming conventional NAS and GCN methods, as well as random search. These findings highlight the significance of a diverse search space and an expressive input representation to enhance the network performance and generalizability.
Paper Structure (23 sections, 9 equations, 5 figures, 8 tables, 1 algorithm)

This paper contains 23 sections, 9 equations, 5 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Skeleton-based HAR
  4. Neural Architecture Search
  5. AutoGCN
  6. Preliminaries
  7. Data representation
  8. Spatial Graph Convolution
  9. Search Space
  10. Architecture
  11. Search Algorithm
  12. Experiments
  13. Datasets
  14. Implementation
  15. Comparison with other SOTA
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Overview of the proposed AutoGCN algorithm.
  • Figure 2: Architecture skeleton of the AutoGCN algorithm.
  • Figure 3: Influence of the rollout parameter on the model performance. The star indicates the highest accuracy achieved in the rollout experiments.
  • Figure 4: Policy values for the best-performing model on the X-Sub dataset. One update cycle contains 20 student architectures.
  • Figure 5: Policy values for the best-performing model on the X-View dataset. One update cycle contains 30 student architectures.