Deep Ensemble CNN Architecture Search for Efficient Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning models face challenges in generalization capability on image processing tasks, requiring manual expertise for neural architecture design and existing automatic methods suffer from low efficiency and poor diversity.
Innovation Solution
A method for generating a deep ensemble model through random sampling of CNN models, constructing a training dataset, training and evaluating CNN models, and using a surrogate model and multi-objective optimization to create a shared block structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual design of neural architecture is used, then model performance can be optimized, but it requires specialized expertise and extensive experience
Solution Approach 1:
The system performs automatic neural architecture design through the neural architecture search module, eliminating the need for manual expert intervention. The algorithm autonomously searches and determines optimal neural network architectures, layer configurations, and hyperparameters, making the process self-service and accessible without specialized expertise.
Solution Approach 2:
The patent replaces the manual mechanical design process with an automated computational system. The neural architecture search module uses algorithms to automatically determine network structures, substituting human expert manual configuration with automated computational optimization.
2Ease of operation
If existing automatic methods are used for designing deep ensemble model architectures, then design accessibility is improved, but efficiency and diversity are poor
Solution Approach 1:
The system performs preliminary actions by pre-defining a comprehensive search space of neural network architectures and pre-training multiple diverse base models. The neural architecture search module is pre-configured with various network structures, and the system pre-generates an ensemble of trained models to serve as building blocks for the deep ensemble architecture, improving subsequent design efficiency.
Solution Approach 2:
The patent segments the deep ensemble model design into distinct components: base model selection, neural architecture search for optimal structure determination, and ensemble construction. This segmentation allows each component to be optimized independently, with the neural architecture search module focusing on structure optimization while the ensemble construction module handles model combination strategies.
3Manufacturing precision
If complex neural architecture is used in deep ensemble model, then model performance is improved, but device complexity increases
Solution Approach 1:
The patent implements a universal neural architecture search module that can handle multiple design objectives and constraints simultaneously. The same search framework is used to optimize various aspects of the ensemble architecture (layer configurations, connection patterns, hyperparameters), making the system multi-functional and reducing overall complexity despite achieving high performance.
Solution Approach 2:
The system optimizes model performance by systematically varying architectural parameters such as network depth, width, and connectivity patterns through the neural architecture search. By controlling and optimizing these parameters methodically rather than using overly complex structures, the system achieves high performance while managing architectural complexity.
Data Source
AI summary
Provided are a method and apparatus for generating a deep ensemble model for image classification, and a computer device. The method includes: randomly sampling, from a search space of classical convolutional neural network (CNN) models for image classification, a neural architecture of a first group of CNN models; collecting representative image samples based on a scenario of an image classification task, marking the collected image samples with classification labels, constructing a training dataset based on the marked image samples, and training the sampled first group of CNN models based on the training dataset and evaluating image classification performance thereof; generating a second group of CNN models based on a surrogate model and the trained first group of CNN models; and constructing a structure of a deep ensemble model for image classification based on the second group of CNN models and a multi-objective optimization strategy.