Recurrent Bayesian Network Construction via Evolutionary CPPN Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing self-organizing network solutions for radio networks require significant human intervention and struggle with efficient resource utilization due to the need for large training datasets and pre-specified network structures, particularly in training recurrent Bayesian networks.
Innovation Solution
A construction system and method for recurrent Bayesian networks using an evolutionary algorithm and fitness function to evolve a population of compositional pattern-producing networks (CPPNs), generating a task network that optimizes radio network configurations automatically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large quantity of training samples are used to train a recurrent Bayesian network, then the network can learn useful patterns from the data, but the training time and computational resources required increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-specifying the network structure before training begins. The recurrent Bayesian network structure is designed in advance based on domain knowledge and problem requirements, which eliminates the need to learn structure from data during training. This preliminary structural configuration allows the network to focus computational resources on learning only the conditional probability parameters from training samples, significantly reducing training time while maintaining learning capability
Solution Approach 2:
The patent segments the network construction process into two distinct phases: structure specification (done beforehand based on domain knowledge) and parameter learning (done during training with minimal computational overhead). This segmentation allows the complex task of network construction to be divided into manageable parts, reducing the overall training burden and time requirement
2Productivity
If the network structure is pre-specified, then the training process becomes more efficient and faster, but the network may not adapt optimally to the specific problem domain
Solution Approach 1:
The patent applies local quality by allowing different parts of the network to have different levels of flexibility. The overall network structure is pre-specified for efficiency, but local adjustments can be made to conditional probability tables and connection strengths based on problem-specific requirements. This enables the network to maintain structural efficiency while adapting locally to domain-specific characteristics
Solution Approach 2:
The patent utilizes parameter changes by adjusting the conditional probability parameters within the pre-specified structure to optimize performance for specific problem domains. While the structure remains fixed for efficiency, the parameters (probabilities and weights) can be tuned and adapted during training to match the characteristics of the specific application, thereby achieving both efficiency and adaptability
3Reliability
If manual operations are used for network management and failure repair, then human expertise can be applied to complex problems, but the operational costs and time consumption increase
Solution Approach 1:
The patent applies self-service by enabling the recurrent Bayesian network to autonomously perform network management tasks such as fault detection, diagnosis, and repair decision-making. The network processes input data through its pre-specified structure and learned parameters to automatically generate outputs for network optimization, reducing the need for manual human intervention while maintaining high-level problem-solving capability through the intelligent inference mechanisms of the Bayesian network
Data Source
AI summary
A construction system for recurrent Bayesian networks, a construction method for recurrent Bayesian networks, a computer-readable recording medium, a non-transitory computer program product, and a radio network control system are provided. The method is performed by a processor, and includes: establishing an initial population, and setting the initial population as a current population; establishing a corresponding recurrent Bayesian network for each CPPN in the current population, to obtain a set of recurrent Bayesian networks corresponding to the current population; evolving the current population by using an evolutionary algorithm and a fitness function to obtain a next population; determining whether a termination condition is met; and repeatedly performing the foregoing steps in response to the termination condition being not met, and selecting a solution network in the current population as the task network based on the fitness function in response to the termination condition being met.