ML Model Tuning for Hardware-Specific Accuracy and Power Tradeoffs

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine-learning models are not effectively tuned to balance accuracy with system performance metrics across different deployment systems due to hardware variations, leading to inefficiencies in power consumption and processing time.

Innovation Solution

A system that modifies machine-learning models based on system performance metrics by adjusting parameters and hyperparameters to reduce power and time consumption while maintaining accuracy, using tuning loss functions and simulation to adapt to specific deployment system hardware.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine-learning models are trained with high accuracy using reference data and loss functions, then model accuracy is improved, but power consumption and processing time increase

Engineering Contradiction:

Improvemodel accuracyVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system modifies model parameters and hyperparameters based on deployment system hardware characteristics. Different data types, precision levels, and computational operations are selected to optimize the balance between accuracy and power consumption for specific deployment environments

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The model tuning process dynamically adapts to different deployment systems by receiving hardware capability information and adjusting model configurations accordingly. The system can modify computational operations, data precision, and model architecture based on real-time hardware capabilities

Inventive Principle:

Principle #15Dynamics

2Measurement precision

If machine-learning models are trained with high accuracy using reference data and loss functions, then model accuracy is improved, but processing time increases

Engineering Contradiction:

Improvemodel accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system modifies model parameters and hyperparameters based on deployment system hardware characteristics. Different data types, precision levels, and computational operations are selected to optimize the balance between accuracy and processing time for specific deployment environments

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

Inventive Principle:

Principle #15Dynamics

3Productivity

If machine-learning models are modified to reduce power and time consumption, then system performance metrics are improved, but accuracy may be compromised

Engineering Contradiction:

Improvesystem efficiencyVSAvoidmodel accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system modifies model parameters andhyperparameters based on deployment system hardware characteristics. Different data types, precision levels, and computational operations are selected to optimize the balance between accuracy and power consumption for specific deployment environments

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system uses feedback from deployment system performance metrics to iteratively adjust model parameters. Hardware capability information and performance data are fed back to refine model configurations, ensuring optimal balance between accuracy and efficiency for each deployment context

Inventive Principle:

Principle #23Feedback

4Ease of manufacture

If machine-learning models are universally trained without hardware-specific tuning, then model development is simplified, but performance varies across different deployment systems

Engineering Contradiction:

Improvemodel development simplicityVSAvoidcross-platform performance consistency

Core Design Contradiction:

Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system modifies model parameters andhyperparameters based on deployment system hardware characteristics. Different data types, precision levels, and computational operations are selected to optimize the balance between accuracy and power consumption for specific deployment environments

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20260004183A1Machine-learning model tuning based on system performance metrics of deployment systems

Publication Date: 2026.01.01 ADVANCED MICRO DEVICES INC

AI summary

To tune a machine-learning model for a deployment system, a machine-learning model training system generates multiple tuning steps for the machine-learning model and generates an accuracy loss sensitivity for each tuning step. Each of the tuning steps indicate a corresponding set of one or more parameters and hyperparameters that reduces the impact of the machine-learning model on the system performance of the deployment system. Based on the accuracy loss sensitivities, the machine-learning model training system selects a tuning step with the least impact on the accuracy of the machine-learning model and modifies the machine-learning model based on the selected tuning step. After also modifying the tuned machine-learning model based on a threshold accuracy, the machine-learning model training system provides the tuned machine-learning model to the deployment system.