Hybrid Neural Network Model with Multi-Valued Mask Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models, such as artificial and spiking neural networks, can only be used independently, limiting their ability to balance high accuracy and low calculation efficiency simultaneously, as they cannot be combined to leverage the advantages of different types of neurons.
Innovation Solution
A neural network model that includes intermediate layers with different types of neurons, obtained through a training process using a multi-valued mask, which combines the advantages of various neural networks by updating the initial neural network and mask based on gradient information and performing multi-value processing to achieve a hybrid model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If different types of neural networks are used independently, then each network can maintain its own characteristics and simplicity, but the system cannot achieve both high accuracy and low calculation efficiency simultaneously
Solution Approach 1:
The patent combines multiple types of neural networks (artificial neural networks and spiking neural networks) into a single hybrid model. Different neuron types are integrated within the same network architecture, allowing the system to leverage the high accuracy of artificial neurons and the energy efficiency of spiking neurons simultaneously, thus resolving the contradiction between accuracy and calculation efficiency
Solution Approach 2:
The hybrid neural network model achieves multi-functionality by incorporating different neuron types that serve different purposes within the same system. The network can dynamically switch between or combine processing modes of different neuron types, enabling it to handle various computational tasks with optimized performance and efficiency
2Adaptability or versatility
If different types of neural networks are used independently, then each network structure remains simple and manageable, but the system lacks adaptability and flexibility
Solution Approach 1:
The hybrid neural network is segmented into different neuron types with distinct functional characteristics. Each neuron type maintains its own simple structure and processing rules, while the overall network gains versatility through the combination of these segmented functional units. This segmentation allows complex adaptability to emerge from simple, manageable components
Solution Approach 2:
The patent introduces a mask mechanism as an intermediary that controls and coordinates the interaction between different neuron types. The mask manages the complexity of integrating multiple neuron types by providing a structured interface, allowing each neuron type to remain relatively simple while achieving coordinated adaptability through the mediating mask
Data Source
AI summary
The present disclosure belongs to the technical field of machine learning. Specifically provided is a neural network model, including at least one intermediate layer including different types of neurons which correspond to different types of neural networks. The neural network model is obtained based on an initial neural network and a multi-valued mask during a training process, and the multi-valued mask is obtained by means of performing multi-value processing on a continuous mask. Further provided are a method for training neural network model, a time sequence data processing method, an electronic device, and a readable medium.