Abstract:We propose dynamical systems trees (DSTs) as a flexible class of models for describing multiple processes that interact via a hierarchy of aggregating parent chains. DSTs extend Kalman filters, hidden Markov models and nonlinear dynamical systems to an interactive group scenario. Various individual processes interact as communities and sub-communities in a tree structure that is unrolled in time. To accommodate nonlinear temporal activity, each individual leaf process is modeled as a dynamical system containing discrete and/or continuous hidden states with discrete and/or Gaussian emissions. Subsequent higher level parent processes act like hidden Markov models and mediate the interaction between leaf processes or between other parent processes in the hierarchy. Aggregator chains are parents of child processes that they combine and mediate, yielding a compact overall parameterization. We provide tractable inference and learning algorithms for arbitrary DST topologies via an efficient structured mean-field algorithm. The diverse applicability of DSTs is demonstrated by experiments on gene expression data and by modeling group behavior in the setting of an American football game.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to construct a flexible and powerful model to describe multiple processes that interact with each other through a hierarchical structure. Specifically, the authors propose **Dynamical Systems Trees (DSTs)** to extend traditional Kalman filters, Hidden Markov Models, and non - linear dynamical systems, enabling them to handle multiple interacting group scenarios. ### Main problems 1. **Modeling of multi - process interactions**: Existing models such as Kalman filters, Hidden Markov Models (HMMs), and non - linear dynamical systems have limitations when dealing with multiple interacting processes. These models are usually only able to handle a single or a small number of processes and cannot effectively capture the interactive behaviors in complex groups. 2. **Hierarchical interactions**: Many systems in the real world (such as player interactions in a football game, gene networks in gene expression data) have a hierarchical interaction structure. Existing models have difficulty effectively representing this hierarchical relationship. 3. **Efficient inference and learning**: For complex dynamical systems, efficient inference and parameter estimation are a challenge. Especially when the system contains a large number of hidden states and complex interactions, the computational complexity will increase significantly. ### Solutions To address the above problems, the paper proposes a new model - **Dynamical Systems Trees (DSTs)**, whose main features include: - **Hierarchical structure**: DSTs organize multiple individual processes through a tree - like structure. Each leaf node represents an independent dynamical system (such as a Switching Linear Dynamical System, SLDS), and high - level nodes act as aggregation processes, mediating the interactions between low - level nodes. - **Flexibility**: DSTs can adapt to different types of dynamical systems, including discrete and continuous hidden states as well as discrete and Gaussian emission distributions. This enables the model to handle multiple application scenarios. - **Efficient inference algorithm**: The paper provides an efficient inference and learning algorithm based on structured mean - field, which is applicable to DSTs of any topological structure. This algorithm simplifies the computational complexity in a recursive manner and improves the operability of the model. ### Application examples The paper verifies the effectiveness of DSTs through two experiments: 1. **Gene expression data**: Use DSTs to model the gene expression levels in cancer cells and demonstrate its advantages in capturing complex interactions between genes. 2. **Football game trajectories**: Model the trajectory data of players in American football games and demonstrate the potential of DSTs in handling multi - agent interaction problems. In conclusion, this paper aims to solve the challenges in multi - process interaction modeling by introducing DSTs, especially providing innovative solutions in terms of hierarchical structure and efficient inference.

Dynamical Systems Trees

Relational State-Space Model for Stochastic Multi-Object Systems

Tree-Structured Recurrent Switching Linear Dynamical Systems for Multi-Scale Modeling

Species, Clusters and the 'Tree of Life': A graph-theoretic perspective

A Systematic Monitoring Framework Utilizing Dynamic Bayesian Network in Plant-wide Processes

Hidden Markov Models derived from Behavior Trees

Hierarchical Bayesian Network Modeling Framework for Large-Scale Process Monitoring and Decision Making.

Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data

A multi-agent model of hierarchical decision dynamics

Discovering group dynamics in coordinated time series via hierarchical recurrent switching-state models

Markov-switching decision trees

Divide, Conquer, Combine Bayesian Decision Tree Sampling

Properties of an N Time-Slice Dynamic Chain Event Graph

Decision making in stochastic extensive form I: Stochastic decision forests

Decentralized digital twins of complex dynamical systems

Dynamic Decision Process Modeling and Relation-line Handling in Distributed Cooperative Modeling System

Probabilistic Models for the (sub)Tree(s) of Life

Dirichlet Fragmentation Processes

Structural learning of simple staged trees

Models and algorithms for skip-free Markov decision processes on trees

Graph Switching Dynamical Systems