Prediction Algorithm Learning via Symmetry Subgroup Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for learning prediction algorithms require large datasets, which is impractical due to exogenous constraints such as memory footprint, execution time, and robustness to input modifications, making it challenging to implement on systems with limited resources.
Innovation Solution
A method that uses a symmetry group and its subgroup to generate learning data, optimizing the prediction algorithm under constraints such as memory footprint, execution time, and robustness, by determining a subgroup and its probability law to reduce the dataset size while maintaining algorithm performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning techniques are used to learn prediction algorithms, then the algorithm can achieve good predictive performance, but a very large dataset is needed which is difficult to obtain in practice
Solution Approach 1:
The patent applies preliminary action by pre-defining symmetry groups and invariance properties before the learning process. These mathematical structures are established in advance to guide the learning algorithm, allowing it to work effectively with smaller datasets by leveraging the predefined symmetry constraints rather than requiring large amounts of data to discover patterns independently
Solution Approach 2:
The patent changes parameters by introducing symmetry group parameters and invariance constraints that modify the learning process. By parameterizing the problem through symmetry groups (such as rotation, translation, or reflection groups) and their associated probability laws, the system can generalize from fewer examples while maintaining predictive accuracy
2Quantity of substance
If data augmentation techniques are applied to increase dataset size, then more learning data can be obtained, but exogenous constraints such as memory footprint and execution time are violated
Solution Approach 1:
The patent extracts and utilizes only the essential symmetry properties from the full symmetry group by determining an optimal subgroup. This extraction process identifies the minimal set of transformations needed to achieve the desired robustness, reducing the computational and memory burden compared to applying the complete symmetry group or traditional data augmentation methods
Solution Approach 2:
The patent applies partial action by using a subgroup of the full symmetry group rather than the entire group. The optimization technique determines the appropriate subset of transformations that provides sufficient data augmentation while respecting memory and execution time constraints, avoiding the excessive action of applying all possible symmetry transformations
3Quantity of substance
If a subgroup of the symmetry group is determined using optimization technique, then the learning can be performed on smaller datasets, but additional computational steps are required
Solution Approach 1:
The optimization of the subgroup and its probability law is performed as a preliminary step before the main learning process. By pre-computing the optimal subgroup that balances dataset size reduction with learning effectiveness, the system avoids repeated optimization during training, thereby reducing the overall learning time despite the additional initial computational step
Data Source
AI summary
A method for learning a prediction algorithm, the learning being implemented by a machine learning technique, the method including reception of a current set of learning data, reception of an invariance property of the prediction of the algorithm with respect to the inputs according to an initial symmetry group endowed with an initial probability law, determination of a subgroup of the initial group endowed with a subgroup probability law and intended to apply transformations to the current set, according to an optimization technique using the current set, the initial group and the initial law under an optimization constraint deduced from predetermined constraints, generation of data using the determined subgroup, the law of the subgroup and the whole current set, and implementation of a learning of the algorithm using the generated data.