Abstract:Emergence, a global property of complex adaptive systems (CASs) constituted by interactive agents, is prevalent in real-world dynamic systems, e.g., network-level traffic congestions. Detecting its formation and evaporation helps to monitor the state of a system, allowing to issue a warning signal for harmful emergent phenomena. Since there is no centralized controller of CAS, detecting emergence based on each agent's local observation is desirable but challenging. Existing works are unable to capture emergence-related spatial patterns, and fail to model the nonlinear relationships among agents. This paper proposes a hierarchical framework with spatio-temporal consistency learning to solve these two problems by learning the system representation and agent representations, respectively. Spatio-temporal encoders composed of spatial and temporal transformers are designed to capture agents' nonlinear relationships and the system's complex evolution. Agents' and the system's representations are learned to preserve the spatio-temporal consistency by minimizing the spatial and temporal dissimilarities in a self-supervised manner in the latent space. Our method achieves more accurate detection than traditional methods and deep learning methods on three datasets with well-known yet hard-to-detect emergent behaviors. Notably, our hierarchical framework is generic in incorporating other deep learning methods for agent-level and system-level detection.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to detect the formation and dissipation of emergence in complex adaptive systems (CASs). Specifically, the paper focuses on how to detect emergence based on the local observations of each agent without a centralized controller, which is very important in dynamic systems in the real world, such as network - level traffic congestion. Existing methods are unable to capture the spatial patterns related to emergence and have difficulty in modeling the nonlinear relationships between agents. Therefore, the paper proposes a new framework - the Hierarchical Framework with Spatio - Temporal Consistency Learning (HSTCL) - to overcome these challenges. ### Background of the Paper Complex adaptive systems (CASs) are composed of autonomous, adaptive, and interacting agents. The interactions of these agents at the micro - level will lead to emergence at the macro - level. Emergence refers to a global property exhibited by the system as a whole, which cannot be simply deduced from the properties of individual agents. For example, when the distance between vehicles is significantly reduced, network - level traffic congestion will occur on the road network, which is an emergence phenomenon. ### Existing Problems 1. **Neglect of Spatial Patterns**: Existing methods such as DETect mainly rely on agent feedback to monitor emergence, but this method may overlook the spatial patterns highly related to emergence. 2. **Assumption of Linear Relationships**: Existing methods usually use linear regression to model the relationships between agents and their neighbors, which may result in the inability to capture nonlinear relationships. ### Solutions The paper proposes a hierarchical framework HSTCL, which solves the above problems in the following ways: 1. **Hierarchical Structure**: - **Agent - level Detection**: Each agent conducts detection based on its local observations and communicates with its neighbors to enhance the detection results. - **Region - level Detection**: The detection results of agents in each region are aggregated to form the region state. - **System - level Detection**: The global monitor analyzes the region states and conducts system - level detection. 2. **Spatio - Temporal Encoders (STEs)**: - **Spatial Transformer**: Used to model the relationships between agents and their neighbors at each time step. - **Time Transformer**: Used to model the changes of agents over time. 3. **Spatio - Temporal Consistency Learning (STCL)**: - **Time Consistency Loss**: Ensures that the representations of agents within the time window are consistent. - **Spatial Consistency Loss**: Ensures that the representations of agents and their neighbors are spatially consistent. ### Main Contributions 1. **Hierarchical Framework**: Captures the spatial patterns related to emergence in a bottom - up manner. 2. **Spatio - Temporal Encoder**: Designs spatial and time transformers that can model the nonlinear relationships between agents and the complex evolution of the system. 3. **Self - supervised Learning**: Learns the representations of agents and the system by minimizing spatio - temporal dissimilarity, avoiding the problems brought by data augmentation and negative samples. 4. **Experimental Verification**: Conducted extensive experiments on three datasets with known but difficult - to - detect emergence phenomena, verifying that HSTCL is superior to existing methods. ### Conclusion HSTCL effectively solves the challenges of detecting emergence in complex adaptive systems through its hierarchical structure and spatio - temporal consistency learning, especially in distributed settings, based on the local observations of agents.

A Hierarchical Framework with Spatio-Temporal Consistency Learning for Emergence Detection in Complex Adaptive Systems

A Hierarchical Framework with Spatio-Temporal Consistency Learning for Emergence Detection in Complex Adaptive Systems

An Unsupervised Framework for Anomaly Detection in a Water Treatment System

Towards Decentralised Detection of Emergence in Complex Adaptive Systems

Hierarchical Learning for Emergence of Social Norms in Networked Multiagent Systems.

Finding emergence in data by maximizing effective information

Decentralised Detection of Emergence in Complex Adaptive Systems

Modeling Spatio-temporal Dynamical Systems with Neural Discrete Learning and Levels-of-Experts

A Hierarchical Approach to Conditional Random Fields for System Anomaly Detection

Scene Classification Based on a Hierarchical Convolutional Sparse Auto-Encoder for High Spatial Resolution Imagery

SACCOS: A Semi-Supervised Framework for Emerging Class Detection and Concept Drift Adaption Over Data Streams

Fast and Robust State Estimation and Tracking via Hierarchical Learning

Learning Hierarchical Dynamics with Spatial Adjacency for Image Enhancement

An Unsupervised Spatiotemporal Graphical Modeling Approach to Anomaly Detection in Distributed CPS

A Unified Replay-based Continuous Learning Framework for Spatio-Temporal Prediction on Streaming Data

A Novel Framework for Spatio-Temporal Prediction of Environmental Data Using Deep Learning

A Deep Learning Framework for Self-evolving Hierarchical Community Detection

Spatio–Temporal Attention-Based Deep Learning Framework for Mesoscale Eddy Trajectory Prediction

Online Intrusion Scenario Discovery and Prediction Based on Hierarchical Temporal Memory (HTM)

An Unsupervised Anomaly Detection Approach Using Energy-Based Spatiotemporal Graphical Modeling

Deep Spatio-Temporal Multiple Domain Fusion Network for Urban Anomalies Detection