Dynamic Neural Distribution Function for Autonomous Class Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques, such as Support Vector Machines (SVM), face challenges in representing entire data sets and adapting to new classes, leading to sub-optimal decision boundaries and inefficient generalization, especially when dealing with non-linearly separable data and new arriving classes.
Innovation Solution
The Dynamic Neural Distribution Function (DNDF) architecture learns and updates individual class distributions independently, using a Cascade Error Projection (CEP) neural network algorithm, allowing for autonomous learning and self-adjustment of neural gains, enabling faster and more efficient classification without competing with other classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If Support Vector Machine (SVM) is used to separate between two classes, then decision boundary can be established with relatively small data samples, but the decision boundary cannot capture the whole set of data points forming the data structure of each class
Solution Approach 1:
The patent segments the decision boundary into multiple neural distribution functions, each representing a different class. Each neural distribution function is trained independently on its corresponding class data, allowing the system to capture the data structure of each class separately. This segmentation enables the system to represent the entire dataset accurately while maintaining computational efficiency.
Solution Approach 2:
The patent transforms the traditional two-class decision boundary problem into a multi-class problem by introducing neural distribution functions that operate in a higher-dimensional space. Each neural distribution function maps class data into a distinct neural space, allowing for more accurate representation of complex data structures while maintaining separability.
2Productivity
If SVM uses only a few data points to represent each class, then computational resources are reduced, but generalization feature becomes insufficient when more data and new types of samples arrive
Solution Approach 1:
The patent implements dynamic learning by allowing neural distribution functions to be trained independently and sequentially. When new data or new classes arrive, the system can train new neural distribution functions without retraining existing ones, enabling continuous adaptation to changing data distributions while maintaining computational efficiency through independent training of individual functions.
Solution Approach 2:
Each neural distribution function trains independently on its own class data without requiring intervention from other classes or the entire dataset. This self-service training mechanism allows the system to automatically adapt to new data and classes while maintaining the efficiency of individual function training, thereby improving generalization capability without sacrificing computational productivity.
3Productivity
If conventional machine learning techniques are used, then classification can be performed, but learning time and computational resources increase when new classes are introduced
Solution Approach 1:
The patent segments the learning process into independent neural distribution functions, each responsible for a specific class. When new classes are introduced, only the corresponding neural distribution function needs to be trained, rather than retraining the entire classification system. This segmentation dramatically reduces learning time and computational resources while maintaining high classification speed for existing classes.
Data Source
AI summary
The present disclosure discusses dynamic supervised learning (DSL) and dynamic neural distribution function (DNDF) machine learning architectures and platforms. In contrast to existing ML approaches, DNDF accommodates a whole data structure via a neural network distribution function from which a decision boundary is born out. In particular, a neural network learning algorithm is used to extract a decision boundary while a neural distribution function is a neural data distribution approach wherein one or more decision boundaries are extracted among various distributions. Other aspects may be described and/or claimed.