Cortical Simulation via Segmented Clock-Driven Neurons
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Simulating large-scale cortical networks with millions of neurons and synapses poses significant computational, communication, and memory challenges due to the vast number of messages required for near real-time performance, exceeding the capacity of current computing platforms.
Innovation Solution
A distributed computing system with N processors simulates neurons and synapses, using a clock-driven synchronous approach for neurons and an event-driven asynchronous approach for synapses, with message aggregation and synchronization techniques to minimize memory usage and communication overhead, allowing for efficient simulation of cortical networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simulations are run at large scale with millions of neurons and synapses, then the computational accuracy and biological realism are improved, but the computation, communication, and memory requirements exceed the capacity of current computing platforms
Solution Approach 1:
The patent divides the large-scale cortical network into multiple processing domains or chunks, where each domain contains a subset of neurons and synapses. This segmentation allows the simulation to be distributed across multiple processors while maintaining the ability to simulate biologically realistic large-scale networks. The memory requirements for each processor are reduced to manageable levels while the aggregate system maintains high simulation accuracy.
2Speed
If the simulation runs in near real-time with high temporal resolution, then the dynamic behavior accuracy is improved, but the computation and communication overhead increases tremendously
Solution Approach 1:
The patent merges multiple communication operations into fewer aggregate messages. Instead of each processor sending individual messages for every neuron-synapse interaction, the system combines communications at domain boundaries, reducing the total number of messages transmitted across the distributed system while maintaining near real-time simulation speed and dynamic behavior accuracy.
Solution Approach 2:
The patent introduces intermediary data structures and communication protocols that mediate between processors. These intermediaries buffer and aggregate communication traffic, reducing the direct communication overhead between processors while maintaining the temporal resolution needed for accurate dynamic behavior simulation.
3Measurement precision
If all synapses are updated every time step to maintain plasticity rules, then the synaptic dynamics accuracy is improved, but the computation and memory requirements scale with the number of synapses which outnumber neurons by a factor of eight thousand
Solution Approach 1:
The patent applies partial updates to synapses based on activity patterns rather than updating all synapses at every time step. By identifying and updating only the synapses that are actively involved in plasticity events (such as those meeting STDP criteria), the system maintains accurate synaptic dynamics while reducing computation and memory requirements by a significant factor compared to universal updates.
Data Source
AI summary
A cortical simulator optimizing the simulation scale and time through computationally efficient simulation of neurons in a clock-driven and synapses in an event-driven fashion, memory efficient representation of simulation state, and communication efficient message exchanges.