This patent falls within the field of Convolutional Neural Networks configured to perform analyses of multivariate time series. It relates to a padding method specifically designed to configure layers of convolutional neural networks, adapted for the analysis of multivariate time series in a two-dimensional input map.
The objectives are:
- Enhanced Time Series Analysis: The developed padding method improves the ability of convolutional neural networks to process time series data, allowing for more effective and precise analysis.
- Cyclic Structure of Data: The method involves projecting the input map onto a cylinder, which enables the consideration of the cyclic structure of the data. This ensures that convolutional operations are conducted appropriately, avoiding information loss that may occur at the edges of the input map.
- Efficiency in Layer Configuration: The padding method facilitates the configuration of the neural network layers, allowing them to better adapt to the input data, thus improving the efficiency of subsequent operations.
- Flexibility in Different Applications: This approach allows the neural network to be utilized in a variety of contexts where multivariate time series analysis is needed, such as finance, climatology, healthcare, among others.
- Improved Result Accuracy: By applying this method, convolutional neural networks can achieve better results in terms of accuracy for predictions and classifications based on the analyzed time series.
This patent offers an innovative solution for the analysis of multivariate time series, utilizing a padding method that maximizes the effectiveness of convolutional neural networks and enhances the accuracy of the obtained results.
Publication:
- WO2021255514A1, December 23, 2021 (filed on 11/27/2020 - PCT/IB2020/061237)
