Multi-Layer Perceptron Architecture For Times Series Forecasting

Organization Name

Inventor(s)

Sercan Omer Arik of San Francisco CA (US)

Nathanael Christian Yoder of San Francisco CA (US)

Multi-Layer Perceptron Architecture For Times Series Forecasting - A simplified explanation of the abstract

This abstract first appeared for US patent application 18417556 titled 'Multi-Layer Perceptron Architecture For Times Series Forecasting

The present disclosure introduces an architecture for time series forecasting based on multi-layer perceptrons (MLPs) that combine linear models with non-linearities.

Time-domain MLPs and feature-domain MLPs are used alternately to perform operations in both domains, leveraging historical and auxiliary data.
The architecture learns temporal patterns and cross-variate information to improve time series forecasts.

Potential Applications:

Financial forecasting
Demand forecasting in supply chain management
Energy consumption prediction in smart grids

Problems Solved:

Improving accuracy of time series forecasts
Leveraging auxiliary data for better predictions

Benefits:

Enhanced forecasting accuracy
Better utilization of historical and auxiliary data
Improved decision-making based on more accurate predictions

Commercial Applications:

Financial institutions
Retail companies
Energy management companies

Questions about Time Series Forecasting: 1. How does the architecture of time-domain and feature-domain MLPs improve forecasting accuracy?

  - The alternation of linear models in different domains allows for learning temporal patterns and leveraging cross-variate information.

2. What are the potential applications of this architecture beyond traditional time series forecasting?

  - This architecture can be applied to various industries for demand forecasting, financial predictions, and energy consumption estimation.

Original Abstract Submitted

The present disclosure provides an architecture for time series forecasting. The architecture is based on multi-layer perceptrons (MLPs), which involve stacking linear models with non-linearities between them. In this architecture, the time-domain MLPs and feature-domain MLPs are used to perform both time-domain and feature-domain operations in a sequential manner, alternating between them. In some examples, auxiliary data is used as input, in addition to historical data. The auxiliary data can include known future data points, as well as static information that does not vary with time. The alternation of time-domain and feature-domain operations using linear models allows the architecture to learn temporal patterns while leveraging cross-variate information to generate more accurate time series forecasts.