Most analog research to date has been performed against a simulator. I believe there is value evolving against a live EAC. Unfortunately, all of the EACs in production today are relatively slow. Evolving against a real EAC introduces a very real time constraints. Using the uEAC USB driver from jEAC, seeding an analog configuration takes about 1-2 seconds. Thus analog-based evolutionary experiments are inherently IO-bound.
Goals
- Demonstrate a simple Artificial Life Simulation
- Investigate how hidden layers from neural networks map to analog
- Integrate GA/GP into jEAC v2.0
Questions
It wouldn't be research if everything made sense!
Genome representation
What does an EAC genome look like? How do I represent types with varying payloads? For the uEAC, an LLA has 27 functions while sources and sinks have 200 values.