How ebsz/srsnn works
Main Components to Include:
Core Model Layer
• Subcomponents:
• Neuron Models (Izhikevich, Lapicque)
• Synapse Models (Basic, Exponential, Bi-exponential)
• Network Representation
• Spike Processing
Connection Set Algebra (CSA) Layer
• This adjacent to the Core Model Layer
• How it interfaces with the network representation
• Mask and operation components
Evolution Framework
• This as a top-level component
• Subcomponents:
• NES (Natural Evolution Strategy)
• SNES (Separable Natural Evolution Strategy)
• Statistics Collection
Tasks Framework
• This as a separate module
• Various task implementations:
• XOR Task
• MNIST Task
• Pole Balancing
• Pattern Recognition
• Time Series
• Lorenz Task
Utility Layer
• This as a supporting layer
• Include:
• Analysis Tools
• Configuration Management
• Data Handling
• Mathematical Utilities
• Random Number Generation
Hierarchical Structure
• Core components at the center
• Supporting modules arranged around the core
• Clear separation between different functional areas
Related Components
• Cluster neuron models together
• Synapse implementations
• Task implementations in a grid
Flow Direction
• Evolution/optimization flow from top to bottom
• Task execution flow from left to right
• Utility access from the periphery
Special Considerations:
Stochastic Nature
• Random number generation components
• Probability distribution flows
• Stochastic processes in the evolution framework
Mathematical Integration
• CSA formalism interfaces
• Mathematical utility connections
• Where optimization algorithms integrate
Monitoring and Analysis
• Analysis pipelines
• Where statistics are collected
• Logging points
• Version information
• External dependencies
• Document key mathematical formulas or algorithms