Accretion Disc[overy] Net Neural Network Emulation Environment
User Stories: As an artist, I would like to represent my interest in neural network emulation so that I can continue research and development in machine intelligence to support artistic endeavors in support of solving world problems.
Constraints/Acceptance:
- The network should be modular:
- The network can be extended and reduced in connectivity, in situ.
- The network can be extended and reduced in function, in situ.
- Sub-networks can be joined as part of the extension/reduction.
- Each network consists of at least one (1) sub-network.
- The network should be develop capable:
- The network and its components should be local loggable for replaying activities; for debugging/tuning.
- The network and its components should be clock lockable for replaying activities; for debugging/tuning.
- The network and its components should be in a language for wide adoption of large developer pools.
- The network should be production capable:
- The network and its components should be clock local.
- The network should emulate the following Neural Components:
- Dendrites
- Axons
- Soma
- Dendritic Spines
- Neurotransmitter Pools
- Neurotramsmitter Depletion
- Other activities deemed necessary (TBD)
- The network should allow communication:
- Sub-networks between different operations teams
- Secured communications
- Low latency communications
- High bandwidth communications
- Generic API Adapters for input and motor control systems
- The network's sub-networks are processed by:
- Emulated Neural Processors "walk the graph" emulating network activity.
- Hardware Neural Processors can be implemented for hardware acceleration.
- Neural Processors execute common runtime routines specified by neural components which are limited by instruction set(s) of the neural processor specification(s).
- The network's sub-networks should track the following geographic information for purpose of temporal-spatial relationships:
- Local Coordinate System
- Global Coordinate System
- Relative Speeds of Information Propagation
- The network's sub-networks should be programmable:
- Neural Modeling Language (NML); serialized human readable network graph (distances and positions included)
- Neural Computer Aided Design and Drafting; a graphical interface for designing, editing, modeling, implementing, operating and troubleshooting network topology.