Neural Network Design Space Reduction via Dataset Characteristics

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Finding an appropriate neural network architecture from a broad design space is a time-consuming process, even when automated, as existing techniques do not effectively consider dataset characteristics to narrow the design space without risking the exclusion of suitable architectures.

Innovation Solution

A design space reduction apparatus and method that acquires original design space information and dataset characteristics to generate a customized design space, narrowing the options based on the dataset characteristics, ensuring that appropriate architectures are retained for faster identification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the design space of neural network architectures is kept broad to ensure all suitable architectures are included, then the reliability of finding appropriate architectures is improved, but the time required to search through the design space increases significantly

Engineering Contradiction:

Improvereliability of finding appropriate architectureVSAvoidtime to find appropriate architecture

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary analysis of dataset characteristics before the main architecture search process. By examining dataset properties upfront and pre-filtering the design space based on these characteristics, the system prepares a narrowed search space in advance, reducing the time required for the subsequent search while ensuring suitable architectures are not excluded

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system changes the parameters of the design space by filtering architectures based on dataset characteristics. This involves adjusting the design space boundaries and constraints dynamically according to the specific dataset being analyzed, transforming the broad original design space into a customized, narrower search space that maintains reliability while reducing search time

Inventive Principle:

Principle #35Parameter changes

2Productivity

If the design space is narrowed to reduce search time, then the productivity of architecture determination is improved, but the risk of excluding suitable architectures increases

Engineering Contradiction:

Improvespeed of architecture determinationVSAvoidretention of suitable architecture

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system dynamically adjusts design space parameters based on dataset characteristics analysis. By changing the constraints and boundaries of the design space according to specific dataset properties, the system narrows the search space in a controlled manner that maintains productivity while ensuring suitable architectures are retained

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system uses dataset characteristics as feedback to guide the design space narrowing process. By continuously analyzing dataset properties and using this information to adjust the search space, the system ensures that narrowing is performed in a way that maintains reliability while improving productivity

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20230385614A1Design space reduction apparatus, control method, and computer-readable storage medium

Publication Date: 2023.11.30 NEC CORP

AI summary

A design space reduction apparatus (2000) acquires original design space information (10) that represents an original design space of an architecture of a target neural network. The design space reduction apparatus (2000) acquires dataset characteristics information (30) that represents characteristics of a target dataset. The target data set is a collection of data to be analyzed by the target neural network. The design space reduction apparatus (2000) generates customized design space information (20) using the original design space information (10) and the dataset characteristics information (30). The customized design space represents a customized design space of the architecture of the target neural network that is narrower than the original design space.