bootstrap slideshow

EEXIST Future Work

Here are some anticipated areas of future work (aka The Wish List)

Directed Evolution

It could be useful to allow user intervention in the evolution process. If the goal is to evolve some sort of deired behavior, the way in which that evolution arrives at a good solution isn't really important. So having a human monitor the behavior of the system of evolving individuals, so they can interceed in the process, selecting particularly interesting looking individuals, or individuals that seem to be moving in the desired direction. Remember, the evolutionary approach is just a tool here: the goal isn't to show that a population of individuals canevolve in a certain way; the goal is to develop a useful initial configuration of the system.

Further Study of Emergent Behaviors

Simply studying the behavior of the system may prove interesting: seeing what sorts of behavior emerge from simply initial states. The system can currently detect how much the system state changes over time, but more sophisticated analysis and detection of the system's behavior could be incorporated, to allow the system to run through many random initial states, while recognizing those cases where the system is doing something that is deemed "interesting."

Light Tracking

Just as an example application: imagine a photon receptor that responds to light by changing the chemical balance in a section of the system. Expose the sensor to a (virtual) light source, and let the chemical balances change. Now tie the balance of other chemicals to the position of the sensing system. Now try to evolve a system that tracks light. This feels less digital than some of the other early experiements.

WISH LIST

The following is a partial list of desired changes and enhancements to the simulated system:

  • better ability to change the scale and granularity of the time- and space-scaling of the simulator
  • continually display the top 5 members of the population during evolution, possibly with the option of interacting with these members to observe their behavior in response to different stimuli
  • extension to 2D and 3D configurations (currently the system is essentially one dimensional)
  • better ability to snapshot and restore system state
  • training vs testing mode: evolve a population, but during evolution, be able to pause the evolution process, and switch to a mode where individuals can be studied, tested, and so on
  • more experiments with mutation
  • exploring the absolute- vs relative-addressing modes
  • pre-stabilization of individuals: some sort of preliminary stim mode that initially drives the individuals until they stabilize, and then uses those pre-stim'd indoviduals in the standard GA
  • smoothing of an individual's genome
  • breeding via blocks of genomes vs. using each chemical tube as a separate genome.

This list will certainly change over time. We've only started to brush the surface of possibilities The challenge is to think beyond the obvious, in order to discover the unanticipated aspect of the system.