Local Surrogate Models for As-a-Service Connectivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current as-a-service (aaS) systems face issues with intermittent connectivity, high latency, and inability to guarantee real-time decisions, making them unsuitable for critical applications unless a local backup is available, which is expensive and slow, and they lack robustness and equal performance in distributed decision-making scenarios.
Innovation Solution
The system monitors input and output data from a cloud platform to generate local models, such as cheap local surrogates (CLS) and artificial intelligence (AI) models, which can mimic the cloud platform's behavior, allowing for robust, safe, and real-time operations even during connectivity issues by switching to local models when needed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a local backup of the aaS platform is used to ensure real-time decisions during connectivity issues, then reliability is improved, but device complexity and cost increase
Solution Approach 1:
The patent creates simplified local copies (surrogate models) of the cloud platform's decision-making logic rather than maintaining full platform backups. These surrogate models are trained to replicate specific decision-making behaviors, providing real-time local inference capability while occupying minimal resources and maintaining low system complexity.
Solution Approach 2:
The patent segments the decision-making capability from the full aaS platform by training specialized surrogate models for specific decision types. This allows the system to maintain only the essential decision-making logic locally rather than replicating the entire platform, reducing complexity while preserving critical real-time functionality.
2Reliability
If a local backup of the aaS platform is used to ensure real-time decisions, then real-time performance is improved, but cost increases
Solution Approach 1:
The patent creates simplified local copies (surrogate models) of the cloud platform's decision-making logic rather than maintaining full platform backups. These surrogate models are trained to replicate specific decision-making behaviors, providing real-time local inference capability while occupying minimal resources and maintaining low system complexity.
Solution Approach 2:
The surrogate models are designed to be lightweight and computationally efficient, consuming minimal local resources. They can be periodically updated or regenerated from the cloud platform without requiring expensive, persistent full-platform backups, effectively using low-cost approximations instead of expensive exact replicas.
3Reliability
If a local copy of the aaS platform is used during connectivity issues, then reliability is improved, but manufacturing precision and performance equality deteriorate
Solution Approach 1:
The patent creates simplified local copies (surrogate models) of the cloud platform's decision-making logic rather than maintaining full platform backups. These surrogate models are trained to replicate specific decision-making behaviors, providing real-time local inference capability while occupying minimal resources and maintaining low system complexity.
Solution Approach 2:
The system continuously monitors the performance gap between surrogate model predictions and actual cloud platform outputs during connected periods. This feedback is used to iteratively refine and update the surrogate models, ensuring they maintain high decision accuracy and closely mimic the cloud platform's behavior even when operating independently.
4Adaptability or versatility
If the aaS platform is used for distributed decision-making with multiple users, then adaptability is improved, but reliability during connectivity issues deteriorates
Solution Approach 1:
The patent trains surrogate models using aggregated data from multiple users to capture diverse decision-making patterns and scenarios. This enables each local surrogate model to generalize across different user contexts and make robust decisions independently, maintaining performance consistency across distributed deployments even during connectivity issues.
Data Source
AI summary
Embodiments for managing an as-a-service (aaS) computing system by a processor are provided. Input provided by at least one computing device to a cloud platform is monitored. Output provided by the cloud platform to the at least one computing device in response to the input provided by the at least one computing device is monitored. At least one local model is generated based on the input provided to the cloud platform and the output provided to the at least one computing device.