What does everyone think about including "hybridized" AI in the OpenAI effort? By hybridized, I mean combining two areas of AI together to form a hybrid - like having a Genetic Algorithm decide the architecture and weights of a Neural Net. I've actually seen this in a couple of places, and experimented with it myself with some interesting effects and bonuses (like no need for training or application where training is essentially impossible).
I am sure other combinations are possible - like combining a GA with a Bayesian Net, or a Neural Net with a Bayesian Net. A Finite State Machine powered Mobile Agent may also fall into this category.
The hybrids could be carried out further still, like a Neural-Genetic Agent. Hmmm ... is that were Neugents(tm) by Computer Associates got there name?
Anyhow, I think that combining the tools would give us something with alot more power. In writing the specifications, we should keep this in mind. Of course, if all of the packages are interoperable as we are planning, then this should be a *snap*.
Speaking of specifications, if anyone would like to give input, feel free. The specs are still under heavy development, so all (serious) input/ideas will be taken into account.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It would be nice if the specifications that have already been determined were available for review. I feel like I don't have any idea where you need suggestions.
--Eli
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I recently wrote a hybrid neural net/ga system as a demonstration of the idea, and also used it to help predict time series. You can find VB6 source code on my site.
www.fuzzgun.btinternet.co.uk
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yeah, I wrote a nn/ga combo that evolved a poker player. At least got better than just a random player :). If I can dig it back up (almost embarassed to show the code), I'll stick it up somewhere.
I think that nn/ga hybrids would be great for games in general. Image quake monsters that fight each other and evolve into smarter opponents (and maybe find flaws in the engine to exploit).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
What does everyone think about including "hybridized" AI in the OpenAI effort? By hybridized, I mean combining two areas of AI together to form a hybrid - like having a Genetic Algorithm decide the architecture and weights of a Neural Net. I've actually seen this in a couple of places, and experimented with it myself with some interesting effects and bonuses (like no need for training or application where training is essentially impossible).
I am sure other combinations are possible - like combining a GA with a Bayesian Net, or a Neural Net with a Bayesian Net. A Finite State Machine powered Mobile Agent may also fall into this category.
The hybrids could be carried out further still, like a Neural-Genetic Agent. Hmmm ... is that were Neugents(tm) by Computer Associates got there name?
Anyhow, I think that combining the tools would give us something with alot more power. In writing the specifications, we should keep this in mind. Of course, if all of the packages are interoperable as we are planning, then this should be a *snap*.
Speaking of specifications, if anyone would like to give input, feel free. The specs are still under heavy development, so all (serious) input/ideas will be taken into account.
It would be nice if the specifications that have already been determined were available for review. I feel like I don't have any idea where you need suggestions.
--Eli
I recently wrote a hybrid neural net/ga system as a demonstration of the idea, and also used it to help predict time series. You can find VB6 source code on my site.
www.fuzzgun.btinternet.co.uk
Yeah, I wrote a nn/ga combo that evolved a poker player. At least got better than just a random player :). If I can dig it back up (almost embarassed to show the code), I'll stick it up somewhere.
I think that nn/ga hybrids would be great for games in general. Image quake monsters that fight each other and evolve into smarter opponents (and maybe find flaws in the engine to exploit).
See http://www.botepidemic.com/neuralbot/