Unsupervised Generative Digital Twin for Complex System Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating complex systems, such as biological cells or weather, face challenges in accurately modeling behavior due to sparse and uncertain data, requiring substantial real-time processing and computation resources, and are unsuited for systems with limited data availability, leading to issues like forgetting and high resource demands.
Innovation Solution
A computer-implemented method for generating a digital twin of a complex system using unsupervised learning to create a manifold representing the variability of the training dataset, allowing for decoupled reinforcement learning without relying on supervised operations, enabling efficient computation and storage gains and flexible adjustment of model complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reinforcement learning models are trained with supervised learning to minimize gaps between predictions and observations, then prediction accuracy is improved, but substantial real-time processing resources and constant interaction with real-world data are required
Solution Approach 1:
The patent applies preliminary action by performing unsupervised learning during an offline training phase to learn the manifold structure of the complex system before actual prediction is needed. This pre-learning of the data distribution and relationships allows the model to make accurate predictions without requiring resource-intensive supervised training at prediction time, thus resolving the contradiction between prediction accuracy and computational resource consumption.
2Reliability
If reinforcement learning models are trained with supervised learning using ground-truth data, then model reliability is improved, but the method is unsuited for systems with sparse or limited data availability
Solution Approach 1:
The patent inverts the traditional supervised learning approach by using unsupervised learning to learn the manifold structure without requiring labeled ground-truth data. Instead of using ground-truth data to supervise the learning process, the model autonomously discovers patterns and relationships in the data distribution, making it suitable for systems with sparse or limited data while maintaining reliability through the learned manifold representation.
3Loss of information
If generative models are used to reconstruct missing data in time series, then data completeness is improved, but substantial computational resources and storage are required
Solution Approach 1:
The patent extracts only the essential manifold structure and data distribution characteristics from the training data during offline unsupervised learning, rather than storing or processing complete datasets. This extraction of core patterns allows the model to reconstruct missing data effectively while requiring minimal storage resources, as only the learned manifold parameters need to be retained rather than the entire dataset.
4Measurement precision
If constant interaction with real-world data is maintained for model updating, then prediction accuracy is improved, but the approach is limited to specific categories of complex systems
Solution Approach 1:
The patent creates a digital twin (copy) of the complex system by learning its manifold structure offline from available data. This digital twin can then be used for predictions and simulations without requiring constant interaction with the real system, making the approach applicable to any complex system for which sufficient training data can be collected offline, thereby expanding system applicability while maintaining prediction accuracy.
Data Source
AI summary
Generating a digital twin of a complex system including receiving at least one training dataset in which each sample includes information on a state and on associated action, including related time information, training a generative model over states, actions and time information to learn a topological space representing attainable system states, in an unsupervised fashion over those states, actions and time information, wherein the generative model learns the mapping to realistic samples includes the space and transitions associated with those samples subject to the actions, and outputting a digital twin including the topological space and transitions between the attainable states subject to the actions, for simulating behaviors of the system by the digital twin to properly achieve one or more tasks pertaining to the system. Applications to reinforcement learning, notably for biological cells.