Simplicial Complex IOV Knowledge Base for Rule-Learning Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a lack of a general basic means applicable to the entire Internet of Vehicles (IOV) field for transitioning from rule-based to learning-based systems, particularly in the integrated mode, where existing solutions focus on specific problems and fail to leverage existing knowledge effectively.
Innovation Solution
A method and apparatus for representing and importing IOV knowledge using a simplicial complex (SC), which involves mapping input variables to output values through a function, representing safe boundaries, and creating a knowledge base by discrete sampling and cell formation, enabling efficient knowledge representation and integration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If transition in split mode is used to completely adopt learning-based systems, then the system can achieve pure learning-based operation, but existing rule-based knowledge is abandoned and wasted
Solution Approach 1:
The patent merges rule-based knowledge systems with learning-based systems by representing rule-based knowledge in a format compatible with neural networks. The knowledge base is transformed into a neural network structure where rules are encoded as weighted connections, allowing both symbolic reasoning and learning to coexist and reinforce each other.
Solution Approach 2:
Instead of discarding rule-based knowledge during transition to learning-based systems, the patent recovers and preserves this knowledge by encoding it into the neural network structure. The rule-based knowledge is not lost but transformed into a form that can be utilized by the learning-based system, maintaining its value while enabling adaptive learning.
2Loss of information
If transition in integrated mode is used to utilize rule-based knowledge in learning-based framework, then knowledge retention is improved, but the form of rule-based system must be abandoned
Solution Approach 1:
The patent creates a copy of the rule-based knowledge structure within the neural network framework. The knowledge base is replicated and transformed into neural network weights and connections, allowing the same knowledge to be accessed both as traditional rules and as learned patterns, thus retaining knowledge while adapting to the new system architecture.
Solution Approach 2:
The patent designs a universal knowledge representation that serves multiple functions: it maintains the structure of rule-based knowledge for interpretability while simultaneously functioning as a neural network for learning and adaptation. This multi-functional design allows the system to leverage both symbolic reasoning and statistical learning without requiring separate systems.
3Reliability
If specific problem-focused attempts are made in integrated mode, then particular issues can be addressed, but a general basic means applicable to entire IOV field is lacking
Solution Approach 1:
The patent creates a universal knowledge base transformation method that can be applied across all IOV scenarios. The approach of encoding rule-based knowledge into neural network structures is designed to be domain-agnostic and can handle various types of knowledge (navigation, control, perception) uniformly, providing both specific problem-solving capability and general applicability.
Solution Approach 2:
The patent segments the overall IOV knowledge base into modular components that can be independently transformed and integrated. Each knowledge domain (e.g., navigation, control, perception) can be processed separately through the transformation method, allowing specific problems to be addressed while maintaining a unified framework applicable to the entire system.
Data Source
Figure 1~3
Figure 4A
Figure 4B
AI summary
An embodiment of the present invention provides a method for representing internet of vehicles IOV knowledge based on a simplicial complex SC. The method includes: representing k'-dimensional knowledge in a k-dimensional continuous space by using a k-dimensional SC with k'-dimensional function values, where a coordinate of a vertex of the SC is values (x1, ..., xk) of k input variables, a function value of the vertex is k' output values (y1, ..., yk') of a function, and a relationship between the coordinate and the function value is (y1, ..., yk') = f (x1, ..., xk'), where f is a mapping function based on IOV knowledge, and k and k' are natural numbers; and representing a safe boundary of the IOV knowledge by using a boundary of the SC, where the IOV knowledge includes a steering wheel angle of an ego-vehicle, a road curvature, a speed of the ego-vehicle, and an inter-parameter relationship compliant with an objective law of vehicle dynamics, where the steering wheel angle of the ego-vehicle, the road curvature, and the speed of the ego-vehicle are obtained by using a sensor on the ego-vehicle.