Dynamic Neural Function Library for Adaptive AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks are based on antiquated models that compromise neuron functions, ignoring temporal activation patterns and feedback, leading to dedicated machines rather than adaptive learning systems.
Innovation Solution
A hierarchical array of digital synapse circuits and neuron soma circuits that simulate analogue values, enabling autonomous learning through Synaptic Time Dependent Plasticity (STDP), allowing synapse strength to change based on temporal differences between input and output pulses, and using a Dynamic Neural Function Library for re-usable and combinable functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional McCulloch-Pitts neurons with static synaptic strength are used, then the network structure is simple and easy to implement, but the network lacks adaptability and cannot perform autonomous learning
Solution Approach 1:
The patent transforms static synaptic strength values into dynamic variables that change over time based on temporal patterns of neural activation. The synaptic strength is no longer a fixed parameter but evolves according to the timing relationships between pre-synaptic and post-synaptic spikes, enabling the network to adapt and learn autonomously while maintaining a relatively simple neural unit structure
Solution Approach 2:
The invention changes the parameter of synaptic strength from a static value to a time-dependent variable. By introducing temporal dynamics into the synaptic strength parameter through STDP rules, the network gains adaptability and learning capability without fundamentally altering the basic neural unit architecture, thus resolving the contradiction between simplicity and adaptability
2Productivity
If static synaptic strength values are used for training, then the training process is straightforward with manual adjustment, but the training time is excessive and the network cannot learn efficiently
Solution Approach 1:
The patent implements self-service learning through autonomous synaptic modification. The network automatically adjusts its own synaptic strengths based on temporal correlation between neural spikes without requiring manual intervention or extensive external training data. This self-organizing capability dramatically reduces training time and improves training efficiency by enabling the network to learn patterns autonomously through its intrinsic temporal dynamics
Solution Approach 2:
The invention introduces feedback mechanisms where the timing of post-synaptic spikes feeds back to modify pre-synaptic synaptic strengths. This temporal feedback loop allows the network to continuously refine its connections based on observed patterns, enabling efficient learning without lengthy manual training processes and reducing the time required to achieve desired network behavior
3Adaptability or versatility
If dedicated machines are created for specific tasks, then the functional results are reliable and consistent, but the system lacks versatility and cannot be reused for different applications
Solution Approach 1:
The patent creates a universal neural network framework that can perform multiple functions through a single adaptable system. By endowing the network with autonomous learning capabilities through temporal plasticity rules, the same network architecture can be applied to different tasks and applications, eliminating the need for dedicated machines while maintaining functional reliability through consistent learning mechanisms across diverse problems
4Adaptability or versatility
If temporal activation patterns and feedback are ignored, then the computational model is simpler and faster to compute, but the biological realism and learning capability are compromised
Solution Approach 1:
The patent segments the complex temporal processing into discrete spike timing events and incremental synaptic updates. Rather than continuously processing temporal patterns, the system processes temporal information through discrete spike events that trigger discrete synaptic strength adjustments, making the computational process more manageable while preserving temporal dynamics and enhancing learning capability
Data Source
AI summary
A method for creating a dynamic neural function library that relates to Artificial Intelligence systems and devices is provided. Within a dynamic neural network (artificial intelligent device), a plurality of control values are autonomously generated during a learning process and thus stored in synaptic registers of the artificial intelligent device that represent a training model of a task or a function learned by the artificial intelligent device. Control Values include, but are not limited to, values that indicate the neurotransmitter level that is present in the synapse, the neurotransmitter type, the connectome, the neuromodulator sensitivity, and other synaptic, dendric delay and axonal delay parameters. These values form collectively a training model. Training models are stored in the dynamic neural function library of the artificial intelligent device. The artificial intelligent device copies the function library to an electronic data processing device memory that is reusable to train another artificial intelligent device.