Multi-branch Neural Network Architecture Search via Block Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic machine learning network architecture search technologies are time-consuming, labor-intensive, and require substantial computational resources, making them inaccessible to small and medium-sized enterprises for applications like industrial defect detection.
Innovation Solution
A multi-branch network architecture searching method and system that parameterize the overall network architecture using block design elements and employ a hyperparameter optimization algorithm to efficiently search for optimal neural network architectures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If existing automatic machine learning network architecture search technologies are used, then network architecture can be automatically searched, but the process is time-consuming and requires substantial computational resources
Solution Approach 1:
The patent segments the network architecture into reusable block design elements (e.g., convolutional blocks, pooling blocks, normalization blocks) that can be combined in different configurations. This segmentation allows the search space to be systematically explored through combinatorial generation rather than exhaustive search, significantly reducing time requirements while maintaining automation.
Solution Approach 2:
The patent employs dynamic hyperparameter optimization where the block design elements and their configurations are adaptively adjusted based on performance metrics. The system dynamically selects and refines architectural components during the search process, enabling efficient exploration of the search space without requiring fixed, time-consuming search protocols.
2Extent of automation
If existing automatic machine learning network architecture search technologies are used, then network architecture can be automatically searched, but substantial computational resources are required
Solution Approach 1:
Instead of exhaustively searching all possible network architectures (excessive action), the patent employs partial action through selective combination of block design elements. The system generates a manageable subset of candidate architectures by combining a limited set of pre-defined blocks with optimized hyperparameters, achieving sufficient performance without the computational burden of complete search.
Solution Approach 2:
The patent changes the parameters of block design elements (such as filter sizes, channel dimensions, layer depths) to create varied network architectures. By parameterizing the search space through hyperparameter optimization rather than structural enumeration, the system reduces computational resource requirements while maintaining the ability to explore diverse architectural possibilities.
3Use of energy by moving object
If manually designed network architectures are used, then computational resources are reduced, but the process is labor-intensive and time-consuming
Solution Approach 1:
The patent implements self-service through automated generation and evaluation of network architectures. The system automatically combines block design elements, optimizes hyperparameters, and evaluates performance without requiring manual intervention in the architectural design process. This eliminates the labor-intensive nature of manual design while maintaining reasonable computational resource usage through efficient search algorithms.
4Measurement precision
If complex network architectures are searched thoroughly, then model accuracy is improved, but the search process becomes more time-consuming
Solution Approach 1:
The patent incorporates feedback mechanisms where performance metrics from evaluating candidate architectures guide the search process. The system uses feedback from validation results to prioritize promising architectural configurations and adjust search strategies, enabling efficient convergence on high-accuracy models without exhaustive search. This feedback-driven approach maintains measurement precision while reducing search time.
Data Source
AI summary
A multi-branch network architecture searching method includes: obtaining a training dataset with a plurality of input data; obtaining block design elements for blocks, wherein the blocks forms an architecture of a neural network, and the blocks are configured to perform a feature extraction on the input data to generate an output data; for each hyperparameter of the neural network, obtaining at least one hyperparameter setting value; inputting the training dataset, block design elements, and the at least one hyperparameter setting value into a hyperparameter optimization algorithm to generate a hyperparameter combination, wherein the hyperparameter combination includes one of the at least one hyperparameter setting corresponding to each hyperparameter; executing the neural network based on the hyperparameter combination and inputting a test dataset to evaluate a model performance of the neural network; and outputting the hyperparameter combination when the model performance reaches a threshold.