Neuromorphic Neural Network Generation via Brain Part Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neuromorphic computing systems focus on simply implementing neurons and synapses similar to the brain, rather than configuring neural networks in an optimal form that closely mimics the brain's structure and function for specific tasks and resource considerations.
Innovation Solution
A method and apparatus for generating neural networks that select and configure brain parts based on requested functions, using a brain information database to create new neural network configuration information, and map it to neuromorphic hardware, adjusting the network to match available resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If neuromorphic computing systems simply implement neurons and synapses similar to the brain, then the implementation complexity is reduced, but the ability to closely mimic the brain's structure and function for specific tasks is worsened
Solution Approach 1:
The patent segments the brain into distinct functional parts (cerebrum, cerebellum, brainstem, etc.) and implements each part separately with its specific neuron and synapse configurations. This allows the system to maintain simplicity in individual implementations while achieving complex brain-like functionality through the combination of segmented parts, each optimized for its specific function.
Solution Approach 2:
The patent applies local quality by configuring different neuron types (excitatory, inhibitory, interneurons) and synapse properties specifically for each brain part based on its functional requirements. For example, the cerebrum uses specific neuron configurations for cognitive functions, while the cerebellum uses different configurations for motor control, thereby mimicking the brain's specialized structure-function relationships.
2Reliability
If a neural network is configured to closely mimic the brain's structure, then the biological fidelity is improved, but the power consumption and resource requirements increase
Solution Approach 1:
The patent extracts and implements only the essential brain parts and neuron-synapse configurations needed for specific tasks rather than replicating the entire brain. For example, if motor control is the task, only the cerebellum and related pathways are implemented, significantly reducing power consumption while maintaining high biological fidelity for the required function.
Solution Approach 2:
The patent implements partial brain functions sufficient for the required task rather than complete brain replication. By providing just enough neural complexity to achieve the desired biological fidelity for specific functions, the system avoids the excessive power consumption that would result from implementing the entire brain while still achieving reliable brain-like behavior for the target application.
3Productivity
If the neural network configuration is customized for each specific task, then the task performance is improved, but the system complexity and configuration time increase
Solution Approach 1:
The patent pre-configures multiple brain part templates with standardized neuron and synapse configurations during system initialization. These pre-configured templates (cerebrum, cerebellum, brainstem, etc.) can be quickly selected and combined for different tasks, avoiding the need to manually configure each neural network from scratch and thereby reducing configuration complexity while maintaining task-optimized performance.
Solution Approach 2:
The patent creates universal brain part configurations that can serve multiple functions. For example, a pre-configured cerebrum template can be used for various cognitive tasks, and a cerebellum template can serve different motor control requirements. This multi-functionality allows the system to achieve customized task performance through combination of universal building blocks, reducing overall configuration complexity.
Data Source
AI summary
Disclosed herein are a neural network generation method for neuromorphic computing and an apparatus for the same. The neural network generation method for neuromorphic computing includes selecting at least one part of a brain corresponding to a neural network function requested to be generated by an application, determining whether an existing neural network corresponding to the at least one part of the brain is present in a neural network database, when it is determined that no existing neural network is present in the neural network database, generating new neural network configuration information corresponding to the part of the brain based on a brain information database, generating a new neural network by mapping the new neural network configuration information to neuromorphic hardware based on the new neural network configuration information, and storing the new neural network in the neural network database.