Complexity-Based Sample Selection for Deep Neural Network Training

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deep neural network training on large-scale datasets is inefficient due to high computing power and energy consumption, primarily because of the large number of parameters and the need for numerous labeled data samples.

Innovation Solution

An energy-efficient sample selection method based on complexity, employing inter-class sampling through reverse diverse self-paced learning and intra-class sampling using a density-based strategy to construct a lightweight dataset for model training, reducing the number of samples required.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If deep neural network training is performed on large-scale datasets, then model performance is improved, but energy consumption increases and training efficiency decreases

Engineering Contradiction:
Improvemodel performanceVSAvoidenergy consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts and removes redundant or low-value samples from the large-scale dataset through two-stage sampling. First, inter-class sampling identifies and removes easy samples that do not contribute significantly to model performance. Second, intra-class sampling further filters samples within each class to retain only those with high information density. This extraction of essential samples reduces the dataset size while maintaining model performance, thereby reducing energy consumption during training.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter of sample selection criteria by introducing a two-stage sampling framework with different selection strategies for each stage. Inter-class sampling uses one criterion to select between classes, while intra-class sampling uses another criterion to select within classes. This parameter change in selection strategy enables more efficient sample selection that reduces training data volume without sacrificing model performance, thus lowering energy consumption.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If deep neural network training is performed on large-scale datasets, then model performance is improved, but training efficiency decreases

Engineering Contradiction:
Improvemodel performanceVSAvoidtraining efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent extracts and removes redundant or low-value samples from the large-scale dataset through two-stage sampling. First, inter-class sampling identifies and removes easy samples that do not contribute significantly to model performance. Second, intra-class sampling further filters samples within each class to retain only those with high information density. This extraction of essential samples reduces the dataset size while maintaining model performance, thereby reducing energy consumption during training.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter of sample selection criteria by introducing a two-stage sampling framework with different selection strategies for each stage. Inter-class sampling uses one criterion to select between classes, while intra-class sampling uses another criterion to select within classes. This parameter change in selection strategy enables more efficient sample selection that reduces training data volume without sacrificing model performance, thus lowering energy consumption.

Inventive Principle:
Principle #35Parameter changes

3Reliability

If the number of labeled data samples is increased, then model performance is improved, but computing power requirements increase

Engineering Contradiction:
Improvemodel performanceVSAvoidcomputing power
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The patent extracts and removes redundant or low-value samples from the large-scale dataset through two-stage sampling. First, inter-class sampling identifies and removes easy samples that do not contribute significantly to model performance. Second, intra-class sampling further filters samples within each class to retain only those with high information density. This extraction of essential samples reduces the dataset size while maintaining model performance, thereby reducing energy consumption during training.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter of sample selection criteria by introducing a two-stage sampling framework with different selection strategies for each stage. Inter-class sampling uses one criterion to select between classes, while intra-class sampling uses another criterion to select within classes. This parameter change in selection strategy enables more efficient sample selection that reduces training data volume without sacrificing model performance, thus lowering energy consumption.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS20230419171A1Energy-efficient sample selection method based on complexity
Publication Date: 2023.12.28 NORTHWESTERN POLYTECHNICAL UNIV

AI summary

The invention discloses an energy-efficient sample selection method based on sample complexity, which performs sample selection on the raw data sets through two stages of inter-class sampling and intra-class sampling, the object is to select representative samples from large-scale data sets, thereby reducing the number of samples used for model training and achieving the object of lightweight training. Compared with the prior art, the invention has the following advantages: the invention proposes an energy-efficient sample selection method based on complexity, selects representative samples from large-scale datasets for efficient model training, and proves that sample complexity and model training strategies have a very important impact on the efficient training of deep neural networks. The invention also solves the problem of low efficiency of model training based on sample complexity and model training strategies, which has certain significance for alleviating the problem of low efficiency of deep learning model training.