Automated Hardware Resource Optimization for Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deep neural network applications require powerful computing resources and have resource-consuming preprocessing steps, but they often run on heterogeneous hardware with variable configurations, especially in shared environments, necessitating dynamic optimization to ensure optimal performance.

Innovation Solution

An automated hardware resource optimization system dynamically configures available computing hardware by determining optimal batch sizes, tuning performance parameters, and optimizing process flows to maximize GPU/TPU/CPU usage, allowing neural network applications to perform efficiently on heterogeneous hardware resources, including local and cloud-based environments.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If neural network applications run on heterogeneous hardware with variable configurations, then hardware versatility and adaptability are improved, but system performance stability and optimization difficulty worsen

Engineering Contradiction:
Improvehardware adaptabilityVSAvoidperformance stability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system dynamically adjusts batch sizes and preprocessing configurations based on real-time hardware performance metrics and available resources. The automated optimization system continuously monitors hardware conditions and adapts processing parameters to maintain stable performance across heterogeneous environments, transforming static configuration into dynamic adaptation.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system automatically modifies key processing parameters such as batch size, preprocessing intensity, and resource allocation based on detected hardware characteristics and performance feedback. By changing these parameters adaptively, the system resolves the contradiction between hardware diversity and performance stability.

Inventive Principle:
Principle #35Parameter changes

2Productivity

If automated optimization systems dynamically configure hardware resources, then processing efficiency and resource utilization are improved, but system complexity and configuration overhead worsen

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidsystem complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The optimization system operates autonomously by automatically detecting hardware characteristics, selecting appropriate batch sizes, and configuring preprocessing parameters without manual intervention. The system monitors its own performance and self-adjusts to optimal settings, reducing operational complexity while maintaining high efficiency.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system implements closed-loop feedback by continuously monitoring processing performance and hardware resource usage, then using this information to automatically adjust configuration parameters. This feedback mechanism enables the system to maintain optimal efficiency while adapting to changing conditions, managing complexity through automated control.

Inventive Principle:
Principle #23Feedback

3Productivity

If batch sizes are increased to improve processing throughput, then productivity is improved, but memory resource consumption and processing stability worsen

Engineering Contradiction:
Improveprocessing throughputVSAvoidmemory consumption
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The system automatically adjusts batch size as a dynamic parameter based on available memory resources and processing requirements. By continuously optimizing this parameter, the system achieves high throughput while preventing excessive memory consumption that would cause instability.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

Instead of using fixed batch sizes, the system dynamically adapts batch size based on real-time monitoring of memory usage and hardware capabilities. This dynamic approach allows the system to maximize throughput when resources are abundant while maintaining stability when resources are constrained.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3923208B1Automated hardware resource optimization
Publication Date: 2024.03.13 DISNEY ENTERPRISES INC
  • EP3923208B1 patent drawingFigure 1
  • EP3923208B1 patent drawingFigure 2
  • EP3923208B1 patent drawingFigure 3A

AI summary

An automated hardware resource optimization system includes a computing platform having a hardware processor and a system memory storing a software code. The hardware processor is configured to execute the software code to identify computing hardware for hosting a neural network based application, determine, based on a first performance parameter of the computing hardware, a batch size for performing data processing using the neural network based application, and tune, using a data batch having the determined batch size, a second performance parameter of the computing hardware to enable substantially continuous loading of its hardware processor memory. The software code also optimizes, based on the determined batch size and the tuned second performance parameter, a process flow for performing the data processing, and generates a configuration file identifying the computing hardware, the neural network based application, the determined batch size, the tuned second performance parameter, and the optimized process flow.