Virtual Hardware Benchmarking for Machine Learning Deployment

Solution Overview

Problem

Building and deploying machine learning models is costly and inefficient due to the need for manual benchmarking on different hardware platforms, often resulting in resource wastage and time-consuming decision-making.

Innovation Solution

A system that virtually executes machine learning models on candidate hardware platforms to predict the best platform for deployment, using a resource prediction twin to simulate performance metrics and select an optimized hardware platform without actual execution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual benchmarking is performed on different hardware platforms to measure machine learning model performance, then the accuracy of performance measurement is improved, but the time consumption and resource consumption increase significantly

Engineering Contradiction:

Improveperformance measurement accuracyVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates a virtualized hardware platform that replicates the performance characteristics of physical hardware without requiring actual physical devices. The system builds a virtual machine model that copies the essential performance metrics (CPU, GPU, memory, storage characteristics) to enable accurate performance prediction through simulation rather than physical benchmarking, thus resolving the contradiction between measurement accuracy and time consumption

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system performs preliminary characterization of hardware platforms by collecting performance data from a diverse set of physical devices and storing it in a database. This pre-characterization allows the system to quickly query and retrieve performance information for virtualized platforms without needing to perform real-time physical benchmarking, reducing time consumption while maintaining measurement accuracy

Inventive Principle:

Principle #10Preliminary action

2Reliability

If multiple hardware platforms are physically benchmarked to determine the best platform for machine learning model deployment, then the reliability of deployment decision is improved, but the cost and resource consumption increase

Engineering Contradiction:

Improvedeployment decision reliabilityVSAvoidresource consumption

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent creates a virtualized hardware platform that replicates the performance characteristics of physical hardware without requiring actual physical devices. The system builds a virtual machine model that copies the essential performance metrics (CPU, GPU, memory, storage characteristics) to enable accurate performance prediction through simulation rather than physical benchmarking, thus resolving the contradiction between measurement accuracy and time consumption

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system introduces a virtualized platform as an intermediary between the machine learning model and physical hardware. This virtual intermediary simulates hardware performance characteristics, allowing reliable deployment decisions to be made without directly interacting with multiple physical hardware platforms, thereby reducing resource consumption while maintaining decision reliability

Inventive Principle:

Principle #24Intermediary (Mediator)

3Ease of operation

If hypothetical scenarios are used to predict model performance on different hardware, then the ease of operation is improved, but the accuracy of performance prediction deteriorates

Engineering Contradiction:

Improveease of performance predictionVSAvoidperformance prediction accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system uses feedback loops where actual performance data from physical hardware benchmarks is continuously collected, stored in a database, and used to train and update the virtualized platform models. This feedback mechanism ensures that the virtual simulations remain accurate by constantly comparing predicted performance against actual measured performance, resolving the contradiction between ease of operation and prediction accuracy

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system performs preliminary characterization of hardware platforms by collecting performance data from a diverse set of physical devices and storing it in a database. This pre-characterization allows the system to quickly query and retrieve performance information for virtualized platforms without needing to perform real-time physical benchmarking, reducing time consumption while maintaining measurement accuracy

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12462075B2Resource prediction system for executing machine learning models

Publication Date: 2025.11.04 ACCENTURE GLOBAL SOLUTIONS LTD

AI summary

A resource prediction system for executing machine learning models and method are provided. The system includes non-transitory memory storing instructions and a processor configured to execute the instructions to obtain input data including a targeted objective and the constraints, select a deployable machine learning model having an evaluation score that meets a predetermined criterion from among candidate machine learning models, virtually execute the deployable machine learning model on each of candidate hardware platforms according to the constraints, generate an assessment report of the virtual performance metrics set of the deployable machine learning model executed on each of the candidate hardware platforms, and select the suggested hardware platform meeting the predetermined criterion from among the candidate hardware platforms.