Neural Network Knowledge Graph Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optimizing results generated by functions in knowledge graphs is challenging, especially for non-linear problems, as existing methods like linear programming are inadequate and hard-coded optimizations are difficult to maintain and execute efficiently.
Innovation Solution
Transforming knowledge graphs into neural networks allows for the optimization of function results using backpropagation techniques, enabling the handling of non-linear problems and dynamic optimization of parameters through non-linear activation functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If linear programming or hard-coded optimization methods are used, then optimization can be performed for simple linear problems, but the system cannot handle non-linear problems and requires difficult maintenance
Solution Approach 1:
The patent replaces traditional mechanical optimization methods (linear programming, hard-coded algorithms) with a neural network-based system. The knowledge graph is transformed into a computational graph that serves as a neural network, enabling the use of gradient-based optimization methods like backpropagation to solve non-linear optimization problems that were previously intractable with linear programming approaches
Solution Approach 2:
The patent changes the fundamental parameters of the optimization system by introducing learnable weight variables and bias terms into the knowledge graph structure. These parameters can be dynamically adjusted through neural network training processes, allowing the system to adapt to non-linear relationships and complex optimization scenarios without requiring manual reconfiguration of the underlying optimization logic
2Reliability
If hard-coded optimizations are implemented, then specific optimization scenarios can be addressed, but the system becomes difficult to maintain and execute efficiently
Solution Approach 1:
The patent implements a self-service optimization system where the neural network automatically learns and adjusts optimization strategies through training on historical data. The system performs self-optimization by minimizing loss functions through backpropagation, eliminating the need for manual intervention to update optimization logic or fix errors in hard-coded optimization routines
Solution Approach 2:
The transformed knowledge graph serves multiple functions: it maintains the original business logic and rule-based computation capabilities while simultaneously providing a framework for neural network-based optimization. This universal structure can handle both traditional forward queries and new backward optimization queries, reducing maintenance burden by consolidating multiple system requirements into a single unified framework
3Productivity
If traditional forward query processing is used, then results can be generated from given inputs, but backward queries to adjust input values for optimization are difficult
Solution Approach 1:
The patent inverts the traditional query processing approach by enabling backward queries that start from a desired output target and work backwards to determine optimal input values. The computational graph structure allows gradient flow in reverse direction through backpropagation, enabling the system to answer optimization questions like 'what input values will produce this target output?' rather than only 'what output does this input produce?'
Solution Approach 2:
The patent introduces feedback mechanisms through loss functions that measure the difference between actual and target outputs. This feedback is propagated back through the computational graph via backpropagation, automatically adjusting input values to reduce the loss and achieve the desired optimization goal, thereby enabling efficient backward query processing
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for optimizing results generated by functions executed using a rule-based knowledge graph. The method generally includes generating a neural network based on a knowledge graph and inputs for performing a function using the knowledge graph. Inputs for the function are received and used to generate a result of the function. A request to optimize the generated result of the function is received. A loss function is generated for the neural network. Generally, the loss function identifies a desired optimization for the function. Values of parameters in the neural network are adjusted to optimize the generated result based on the generated loss function, and the adjusted values of the parameters in the neural network are output in response to the request to optimize the generated result of the function.