Serverless Resource Allocation for Evolutionary Optimization Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Evolutionary machine learning and optimization models face challenges in accessing and maintaining scalable computational resources, leading to high initial capital investments and operational expenses, which hinder their widespread adoption in various sectors.
Innovation Solution
Utilizing cloud-based serverless services for computational resources that are charged on a pay-as-you-go basis, allowing users to specify exact resource quantities and qualities without upfront investments, and enabling automatic termination of resources after each function evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If traditional computational resources are used for evolutionary machine learning models, then computational power is provided, but high initial capital investments and operational expenses are required
Solution Approach 1:
The patent applies serverless computing where computational resources are allocated temporarily for specific function executions and automatically terminated afterward. This eliminates the need for maintaining permanent expensive infrastructure, converting fixed capital investments into variable operational costs that are charged only when computation is actually needed.
Solution Approach 2:
The patent uses a serverless cloud platform as an intermediary between the user's evolutionary machine learning models and the physical computational hardware. This intermediary abstracts away the complexity of resource provisioning, maintenance, and scaling, providing computational power on-demand without requiring direct management of underlying infrastructure.
2Productivity
If computational resources are allocated for evolutionary optimization, then optimization can be performed, but resource maintenance and provisioning complexity increases
Solution Approach 1:
The serverless computing platform automatically handles resource provisioning, allocation, and termination based on function execution requirements. The system self-manages the computational resources, eliminating the need for users to manually provision servers, configure networks, or maintain infrastructure while still enabling complex evolutionary optimization tasks.
Solution Approach 2:
The patent extracts the resource management complexity from the user's optimization workflow by implementing serverless architecture. The infrastructure management layer is separated and handled by the cloud provider, allowing users to focus solely on their evolutionary machine learning models without dealing with underlying resource provisioning and maintenance.
3Reliability
If fixed computational resources are provided, then stable computation is available, but flexibility to scale resources is limited
Solution Approach 1:
The serverless computing architecture provides dynamic resource allocation that automatically adjusts based on the computational requirements of each function execution. Resources are provisioned only when needed and scaled down afterward, providing both stability during execution and flexibility in resource consumption. This dynamic model allows the system to adapt to varying workloads without requiring over-provisioned fixed infrastructure.
Data Source
AI summary
A request is received from a user compute device to solve an optimization problem. A set of subpopulations is identified from an identified population included in the optimization problem. Each subpopulation from the set of subpopulations is associated with a unique sub-machine resource from a set of sub-machine resources at a set of remote compute devices. Each sub-machine resource is caused to determine an associated set of candidates from an associated subpopulation from the set of subpopulations and for that sub-machine resource. Each sub-machine resource is caused to generate an associated set of offspring based on an objective function associated with the associated set of candidates for that sub-machine resource. Each sub-machine resource is caused to mutate the associated set of offspring to generate an associated set of mutated offspring. An updated population is determined, and a representation of a response is determined based on the updated population.