Dynamic Boltzmann Machine Predicting Time-Varying Second Moments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dynamic Boltzmann machines face challenges in adapting to real-number time-series data due to their binary data handling limitations, making it difficult to effectively predict higher-order moments in time-series datasets.
Innovation Solution
Employing a dynamic Boltzmann machine (DyBM) in conjunction with a generalized auto-regressive heteroscedasticity (GARCH) model to predict time-varying second-order moments of time-series datasets by learning a time-series generative model with eligibility traces and acquiring parameters for accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a dynamic Boltzmann machine uses binary data handling, then the model structure remains simple, but it cannot effectively process real-number time-series data
Solution Approach 1:
The patent transforms the binary data constraint into a continuous value framework by changing the parameter representation. Instead of binary states, the system uses continuous activation values that can represent real-number time-series data, allowing the Boltzmann machine to adapt to financial and other continuous data while maintaining the core probabilistic framework
Solution Approach 2:
The patent introduces dynamic elements by making the network weights and activation values time-dependent. The weights are updated dynamically through learning processes that adapt to temporal patterns in the data, and the activation values evolve over time to capture dynamic relationships in time-series data
2Measurement precision
If a Boltzmann machine learns time-series data, then predictive capability improves, but computational complexity increases
Solution Approach 1:
The patent segments the learning process into distinct components: eligibility trace computation, weight update calculations, and parameter optimization. By dividing the complex learning task into manageable segments with specific functions, the system achieves accurate prediction of higher-order moments while making the computational process more tractable and efficient
3Productivity
If the DyBM learns with eligibility traces, then learning efficiency improves, but the computational overhead increases
Solution Approach 1:
The patent implements preliminary action by pre-computing and storing eligibility traces during the forward pass before the weight update step. This preliminary computation of trace values allows the learning algorithm to efficiently utilize historical information during parameter updates, improving learning efficiency while organizing computational work in advance to reduce overall overhead
Data Source
AI summary
A computer-implemented method includes employing a dynamic Boltzmann machine (DyBM) to predict a higher-order moment of time-series datasets. The method further includes acquiring the time-series datasets transmitted from a source node to a destination node of a neural network including a plurality of nodes, learning, by the processor, a time-series generative model based on the DyBM with eligibility traces, and obtaining, by the processor, parameters of a generalized auto-regressive heteroscedasticity (GARCH) model to predict a time-varying second-order moment of the times-series datasets.