Boundary Graph Machine Learning Algorithm for Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning algorithms are slow to train, require large datasets, and perform poorly with small training examples, often learning unnaturally and becoming slower as they handle larger datasets, which is inefficient and impractical for real-time applications.
Innovation Solution
The boundary graph machine learning algorithm, which constructs a network of nodes connected by 'node boundaries' based on similarity in inputs and differences in outputs, allows for efficient training and testing by determining node outputs through 'walks' along these boundaries, enabling faster learning and improved performance with smaller datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current leading machine learning algorithms (neural networks, SVMs, KNN) are used to achieve high accuracy, then classification and regression performance is improved, but training time increases to several CPU-years requiring clusters and GPU acceleration
Solution Approach 1:
The patent segments the training process into two distinct phases: (1) constructing a boundary graph structure from training data that captures decision boundaries, and (2) performing rapid classification by walking along pre-computed boundaries. This segmentation allows the computationally intensive work to be done once during graph construction, while classification queries are answered quickly by traversing the pre-processed structure.
Solution Approach 2:
The patent performs preliminary action by pre-computing and storing the boundary graph structure during an offline training phase. The boundary graph captures all necessary decision boundary information in advance, so that during online classification, the system only needs to traverse pre-computed paths rather than re-evaluate complex models from scratch.
2Measurement precision
If large training corpora are presented to achieve better learning, then model performance is improved, but training time increases and the system becomes progressively slower to respond
Solution Approach 1:
The boundary graph structure segments the feature space into regions defined by decision boundaries, allowing the system to handle large training corpora efficiently by organizing them into a structured representation. This segmentation enables the system to process more data without proportionally increasing response time, as the boundary graph provides a compact organization of the learned knowledge.
3Reliability
If standard machine learning algorithms are used to handle large datasets, then comprehensive learning is achieved, but the algorithms become progressively slower to respond to test queries
Solution Approach 1:
The patent performs preliminary action by pre-computing the boundary graph structure that encodes all learned decision boundaries. During test queries, the system traverses this pre-computed structure rather than re-evaluating the entire model, enabling fast response times even after learning from large datasets. The heavy computational work is done in advance during graph construction.
4Measurement precision
If large batches of training examples are presented to achieve better learning, then model accuracy is improved, but the training process becomes less adaptable to small datasets and real-time applications
Solution Approach 1:
The boundary graph construction is a dynamic process that adapts to the amount of available training data. The graph structure is built incrementally by processing training examples and updating boundaries, allowing the system to work effectively with small datasets by constructing a simpler graph, or with large datasets by creating a more detailed graph. This dynamic adaptation enables versatility across different data sizes.
Data Source
AI summary
There is provided a system and method for training and utilizing a boundary graph machine learning algorithm. The system including a processor configured to receive a plurality of entry nodes, each of the plurality of entry nodes including an entry node input and an entry node output, add each of the plurality of entry nodes to a graph using the entry node input and the entry node output, receiving a plurality of training nodes, each of the plurality of training nodes including a training node input and a training node output, add each of the plurality of training nodes to the graph when the training node input for each of the plurality of training nodes is similar to the training node output of a closest node and the training node output of each of the plurality of training nodes is different than the training node output of the closest node.