Learning Apparatus Chronological Data Segmentation for Model Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for training models, such as SVM and DNN, face challenges in improving model accuracy due to inadequate dynamic adjustment of learning data based on hyperparameter values, leading to potential failures in accuracy enhancement.
Innovation Solution
A learning apparatus that divides learning data features into sets in chronological order, selects sets for training, and trains models using older data first, optimizing hyperparameters and training methods to enhance model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If learning data is dynamically changed according to hyperparameter values, then the training process becomes more flexible, but model accuracy improvement cannot be ensured
Solution Approach 1:
The patent segments learning data into multiple time series data sets with different time spans (e.g., recent data, historical data). By dividing the learning data into chronological segments and selectively training on different segments based on hyperparameter values, the system achieves both adaptability in training dynamics and controlled accuracy improvement through systematic data selection.
2Device complexity
If learning data is merely changed according to hyperparameter values, then the training process is simplified, but accuracy improvement fails when hyperparameter values are not appropriate
Solution Approach 1:
The patent implements dynamic selection of time series data sets based on hyperparameter values. The training process adaptively changes which time span of data is used (recent vs. historical) according to the hyperparameter settings, creating a dynamic training system that can respond to different configuration scenarios while maintaining accuracy through purposeful data selection.
3Manufacturing precision
If all learning data sets are used for training, then more features are learned, but training time increases and efficiency decreases
Solution Approach 1:
The patent applies partial action by selectively training on specific time series data sets rather than using all available data. Based on hyperparameter values, the system trains on appropriate portions of the data (e.g., only recent data, or only historical data, or a combination), achieving sufficient feature learning without the computational overhead of processing the entire data set.
Data Source
AI summary
Improving accuracy of a model.A learning apparatus according to the present application includes: a dividing unit that divides predetermined learning data features of which are to be learned by a model by training, into a plurality of sets in chronological order; a selection unit that selects sets to be used for the training of the model, from among the sets obtained by the division by the dividing unit; and a training unit that trains the model to learn the features of the learning data included in each of the sets selected by the selection unit, by using the sets in order from the set in which the learning data included is older in time series, among the sets selected by the selection unit.