Enhancing Computational Efficiency and Accuracy in Complex Systems Modeling

Advances in Computational Modeling and Machine Learning

Recent developments in the field of computational modeling and machine learning have shown significant advancements, particularly in the integration of complex systems with data-driven approaches. The focus has been on enhancing the efficiency and accuracy of simulations, especially in dynamic and high-dimensional systems such as epidemics, fluid dynamics, and ecological models. Key innovations include the use of Graph Neural Networks (GNNs) for modeling complex interactions, the development of adaptive online smoothers for real-time data assimilation, and the integration of mechanistic models with deep learning for improved forecasting.

Noteworthy Developments:

  • Integrated Epidemic Simulation Workflow: Combines stochastic agent-based simulation with network-based optimization for submodular intervention strategies, enhancing the efficacy of epidemic response measures.
  • Adaptive Online Smoother: Introduces an information-theoretic lag selection method for smoother-based data assimilation, significantly reducing computational storage demands in high-dimensional systems.
  • Deep Reinforcement Learning for Digital Twins: Proposes a temporal model for complex networked systems, demonstrating improved epidemic resilience through cooperative node strategies.
  • Ergodic Theoretic Approach to Generalization: Provides a theoretical framework for assessing the statistical accuracy of learned dynamical systems, addressing limitations in conventional generalization metrics.
  • Physics-based Graph Neural Networks: Investigates the potential of PhysGNN for predicting breast deformation during mammographic compression, offering a balance between accuracy and computational efficiency.
  • Graph Neural Network Surrogates for Pandemic Response: Combines mechanistic models with GNNs for rapid, on-the-fly adaptations in disease dynamics modeling, enabling timely public health decisions.
  • Universal Neural Symbolic Regression: Develops a tool for learning interpretable network dynamics, demonstrating effectiveness across various scientific domains.
  • Multi-Phase Physics-Informed Neural Networks: Introduces a hybrid method for epidemic forecasting, achieving superior performance by integrating mechanistic insights with neural network flexibility.
  • Scientific Machine Learning in Ecological Systems: Applies Neural ODEs and UDEs to predator-prey dynamics, highlighting the robustness of UDEs in noisy data scenarios.
  • Buffer-Free Online Learning Framework: Introduces ODEStream for streaming time series forecasting, addressing irregularity and concept drift without the need for buffering.
  • Graph Neural Networks for Quiver Mutation: Uses GNNs to characterize quiver mutation classes, uncovering new criteria for mutation equivalence.
  • Flow Reconstruction with Graph Neural Networks: Presents a GACN for reconstructing flow fields from sparse data in time-varying geometries, outperforming conventional methods in accuracy and scalability.
  • Neural Graph Simulator: Introduces the NGS for simulating complex systems on graphs, offering significant computational advantages and real-world application potential.
  • Real-time Simulation of Industrial Particulate Flows: Proposes NeuralDEM for replacing numerical DEM routines with deep learning surrogates, enabling faster and scalable simulations in industrial scenarios.

Sources

An Integrated Epidemic Simulation Workflow for Submodular Intervention Strategies

An Adaptive Online Smoother with Closed-Form Solutions and Information-Theoretic Lag Selection for Conditional Gaussian Nonlinear Systems

Deep Reinforcement Learning for Digital Twin-Oriented Complex Networked Systems

When are dynamical systems learned from time series data statistically accurate?

Graph Neural Networks for modelling breast biomechanical compression

Towards Graph Neural Network Surrogates Leveraging Mechanistic Expert Knowledge for Pandemic Response

Learning Interpretable Network Dynamics via Universal Neural Symbolic Regression

MP-PINN: A Multi-Phase Physics-Informed Neural Network for Epidemic Forecasting

Scientific machine learning in ecological systems: A study on the predator-prey dynamics

ODEStream: A Buffer-Free Online Learning Framework with ODE-based Adaptor for Streaming Time Series Forecasting

Machines and Mathematical Mutations: Using GNNs to Characterize Quiver Mutation Classes

Flow reconstruction in time-varying geometries using graph neural networks

Neural Graph Simulator for Complex Systems

NeuralDEM - Real-time Simulation of Industrial Particulate Flows

Built with on top of