Learning to Communicate with Deep Multi-Agent Reinforcement Learning | Hacker News Reader

Learning to Communicate with Deep Multi-Agent Reinforcement Learning

github.com

73 pointsseycombi5 comments

I think it's really cool that these kinds of agent models are catching on in the RL world, with this, and OpenAI's recent paper. And with https://arxiv.org/abs/1703.03864 makes me wonder how close we are to seeing deep learning resurrecting good ol' A-life. If I may divert for a moment to wistfully recall an old project...

When I was a teen I read Steven Levy's book "Artifical Life". It captured my imagination in a way that the biomorph program in a Blind Watchmaker also had. Ever since then I've played with toy alife models. Some of the attraction is just aesthetic - it's fun to set these things up and design the rules, kinda like a higher-order god-game. Another is the enticing fantasy that something like this will one day lead to AGI. A third is the occasionally fulfilled desire to see the fabled emergent behavior of evolutionary computations, where the agents learn to do something you never expected, or exploit some funky loophole to their advantage.

When I was an undergrad I created an alife system as an honors project. It was an ambitious project, with sexually or asexually reproducing agents controlled by recurrent neural nets. The agents had perception and the ability to move, reproduce, attack, co-operate, or interact with objects. It had an entity framework to make it possible to design things like programmable Skinner boxes, traps, feeders, it had feeding schedules, it had live visualization and a slick Qt interface to edit the entities inline or take control of specific agents and visualize their brains.

I was so proud of it.

My assigned supervisor, who was an austere Georgian category theorist, had zero interest. He compared it unfavorably to a virus. Anyway, it was so dispiriting an experience that I kind of swore off Alife after that. But the fun I had made me realize I didn't want to be a mathematician, so I suppose it played a useful role in my life after all.

I am sorry your supervisor was a jerk, your project sounds cool! And even if it weren't cool, it was his job to make sure you stay motivated and keep learning. He clearly failed catastrophically.

With latest developments in ML/AI space this is a very interesting period to be living in. A few decades later they will probably say: "remember when there was no artificial agents and the computers were those dumb boxes?" I hope you find your way back and join the fun!

That sounds pretty freaking good to me...reminds me of the rtNEAT evolutionary algorithm, which combines genetic algorithms real time to evolve anns.

This type of approach is what led me to the current data model of ibGib. Each ibGib address is a node and will inevitably compete for attention "resources". The interesting aspect is when you think of any agent's snapshot in time as being the source of a fork which could itself be an agent. So each point in time of an agent configuration is itself its own possible branching agent configuration in thr future.

It is monotonically increasing because you never know which agent's state will come in handy to get out of a local max/min. Also, it reuses agent dna (pointers are cheap) so overall this is not as expensive as it sounds.

And worst case is that if the overall data store is strained for storage resources, any individual ibgib has its dependencies. So you can do a dependency graph projection, essentially creating a slimmer copy of the original.

Btw Your username reminds me of a book series I read as a kid...Taliesin was King Arthur's father maybe?

Reminds me of this: https://fatiherikli.github.io/language-evolution-simulation/

That readme is quite appropriate. That is just fun to watch.

Reply on news.ycombinator.com