Staged Neural Network Chain for Time Series Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and volume of data in scientific and technical analyses require advanced mechanisms for organization, reproducibility, and accountability, particularly in distributed systems where collaboration and reuse of data and analysis routines are desired, but existing technologies lack efficient methods for overseeing reproducibility and error checking across large datasets.
Innovation Solution
A neural network chain is trained using a processor to perform analytical functions, where each neural network in the chain receives external inputs and outputs from preceding networks, allowing for sequential training and improved time series predictions, with mechanisms for monitoring accuracy and adapting the chain based on performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single neural network is used to perform the entire analytical function, then the system is simpler to implement, but the prediction accuracy for complex time series data deteriorates
Solution Approach 1:
The patent divides a single complex neural network into multiple smaller neural networks arranged in a chain architecture. Each neural network in the chain performs a portion of the analytical function, with the head neural network processing initial inputs and passing results to subsequent networks, ultimately producing the tail output. This segmentation allows each component network to be simpler while collectively achieving higher prediction accuracy for complex time series data.
2Productivity
If the neural network chain is trained using all available data at once, then training is faster, but the ability to adapt to new patterns deteriorates
Solution Approach 1:
The patent implements staged training where neural networks are trained sequentially in a predetermined order from head to tail. Each neural network is trained before the next one in the chain, allowing the system to build predictive capabilities progressively. This preliminary staged approach enables faster initial training while maintaining adaptability, as each network learns from the outputs of previous networks and can be independently refined.
3Measurement precision
If more neural networks are added to the chain to improve prediction accuracy, then the prediction performance improves, but the training time increases
Solution Approach 1:
The patent reduces overall training time by training neural networks in the chain in a predetermined sequential order rather than requiring iterative training of the entire chain. Each neural network is trained once in sequence from head to tail, eliminating redundant retraining of previous networks when adding new networks to the chain. This preliminary sequential training approach significantly reduces the time loss associated with increasing the number of networks.
Data Source
AI summary
An apparatus includes a processor to: train a first neural network of a chain to generate first configuration data including first trained parameters, wherein the chain performs an analytical function generating a set of output values from a set of input values, each neural network has inputs to receive the set of input values and outputs to output a portion of the set of output values, and the neural networks are ordered from the first at the head to a last neural network at the tail, and are interconnected so that each neural network additionally receives the outputs of a preceding neural network; train, using the first configuration data, a next neural network in the chain ordering to generate next configuration data including next trained parameters; and use at least the first and next configuration data and data indicating the interconnections to instantiate the chain to perform the analytical function.