Machine Learning Architecture Search With Proxy-Based Evaluation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing ML model design processes are time-consuming, resource-intensive, and lack comprehensive systems for optimizing architectures across multiple domains and hardware platforms, often resulting in suboptimal performance due to limited applicability and reliance on narrow performance metrics.

Innovation Solution

A comprehensive ML architecture search system that efficiently discovers optimal ML architectures for specified AI/ML domains and hardware platforms by utilizing a framework that includes a user interface, performance metrics, and a repository, employing evolutionary algorithms and proxy functions to reduce computation time and resource consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional ML model design techniques are used, then model accuracy can be achieved through manual design, but the process is time-consuming and resource-intensive

Engineering Contradiction:

Improvemodel design speedVSAvoidtraining and validation time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent uses proxy functions to create simplified copies of the full training process. These proxies train much faster and use fewer resources, allowing rapid iteration and evaluation of multiple model architectures without requiring full training for each candidate, thus dramatically reducing the time and computational resources needed for model design

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system employs evolutionary algorithms that automatically search for optimal model architectures without manual intervention. The process self-manages the iterative cycle of training, validation, and parameter adjustment, eliminating the need for manual model design and significantly accelerating the overall process

Inventive Principle:

Principle #25Self-service

2Ease of operation

If Neural Architecture Search (NAS) algorithms are used to automatically discover ideal models, then manual design time is reduced, but the process becomes even more computationally intensive and time-consuming

Engineering Contradiction:

Improveautomatic architecture discoveryVSAvoidcomputational resource consumption

Core Design Contradiction:

Ease of operationVSUse of energy by stationary object

Solution Approach 1:

The patent introduces proxy functions that create simplified representations of the full training process. These proxies enable the NAS algorithm to evaluate and compare different architectures using minimal computational resources, allowing automatic architecture discovery without the prohibitive energy consumption of training full models for each candidate

Inventive Principle:

Principle #26Copying

Solution Approach 2:

Instead of training complete models for all candidate architectures, the system uses partial training through proxy functions that provide sufficient information to evaluate architectural quality. This partial action approach maintains the automatic discovery capability while dramatically reducing the computational energy required

Inventive Principle:

Principle #16Partial or excessive action

3Loss of time

If proxy functions are used to predict model accuracy instead of full training, then computation time is reduced, but the correlation between proxy scores and actual performance becomes unreliable

Engineering Contradiction:

Improvetraining timeVSAvoidaccuracy prediction precision

Core Design Contradiction:

Loss of timeVSMeasurement precision

Solution Approach 1:

The patent implements a feedback mechanism where proxy function results are used to guide the evolutionary algorithm, and the best candidates are then evaluated with full training. This feedback loop allows the system to leverage the speed of proxies for initial screening while maintaining the precision of full training for final validation, balancing both time efficiency and measurement accuracy

Inventive Principle:

Principle #23Feedback

4Adaptability or versatility

If existing NAS solutions are used, then architecture search can be performed, but they are limited to specific AI/ML domains and performance metrics

Engineering Contradiction:

Improveapplicability to multiple domainsVSAvoidsystem complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent designs a universal framework that can handle multiple AI/ML domains and performance metrics through a single unified system. The evolutionary algorithm and proxy functions are configured to work across different domains by adjusting the fitness function and training procedures, eliminating the need for separate specialized systems for each domain while maintaining comprehensive adaptability

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12367249B2Framework for optimization of machine learning architectures

Publication Date: 2025.07.22 INTEL CORP

AI summary

The present disclosure is related to framework for automatically and efficiently finding machine learning (ML) architectures that are optimized to one or more specified performance metrics and/or hardware platforms. This framework provides ML architectures that are applicable to specified ML domains and are optimized for specified hardware platforms in significantly less time than could be done manually and in less time than existing ML model searching techniques. Furthermore, a user interface is provided that allows a user to search for different ML architectures based on modified search parameters, such as different hardware platform aspects and/or performance metrics. Other embodiments may be described and/or claimed.