Slashdot Mirror

← Back to Stories (view on slashdot.org)

Incorporating Swarm Intelligence Into Computer AI

Posted by Soulskill on from the don't-show-this-to-the-left-4-dead-guys dept.

An anonymous reader writes "From optimizing truck delivery routes to inspiring nerve-cell-based cognition models, ant intelligence has arrived. From the Economist: 'In 1992 Dr. Dorigo and his group began developing Ant Colony Optimisation (ACO), an algorithm that looks for solutions to a problem by simulating a group of ants wandering over an area and laying down pheromones. ACO proved good at solving travelling-salesman-type problems. Since then it has grown into a whole family of algorithms, which have been applied to many practical questions. ... Ant-like algorithms have also been applied to the problem of routing information through communication networks. Dr. Dorigo and Gianni Di Caro, another researcher at IDSIA, have developed AntNet, a routing protocol in which packets of information hop from node to node, leaving a trace that signals the "quality" of their trip as they do so. Other packets sniff the trails thus created and choose accordingly. In computer simulations and tests on small-scale networks, AntNet has been shown to outperform existing routing protocols."

10 of 64 comments (clear)

  1. The Ant Networking Protocol was a huge success by cyber-vandal · · Score: 4, Funny

    Until it passed by a group of people having a picnic.

  2. Re:Hill Climbing by caffeinemessiah · · Score: 5, Insightful

    It just sounds like the classic hill climbing algorithm to me.

    That's because it's very similar -- with a massive stochastic component. It might be effective at routing, but I image leaving "pheromone traces" over network routes to indicate quality (latency, bandwidth, whatever) is something that will make sure security researchers have jobs for a long, long time.

    --
    An old-timer with old-timey ideas.
  3. Nothing new here... by Anonymous Coward · · Score: 4, Informative

    Dorigo et al. made their groundbreaking paper in 1996 (based on observations by Deneubourg in 89), and then nothing. Nothing new on the theoretical part, no new application. Even routing using ACO like algorithm has been published as early as 1994. The newest extension based on these algorithms is an interactive distributed image retrieval system by Picard et al. back in about 2006 (as far as I know). So nothing new nor groundbreaking here.

  4. Re:Swarm UDP? by bhcompy · · Score: 4, Funny

    Those blast marks on that router are too precise for sand people

  5. Interesting Paper by stoanhart · · Score: 3, Informative

    I just covered Ant-based load balancing on communications networks in a distributed systems class. Here's the paper we read. It's an easy read, and quite interesting.

    http://www.hpl.hp.com/techreports/96/HPL-96-76.pdf

    1. Re:Interesting Paper by dikarus · · Score: 3, Informative

      that was one of the original papers on the topic, I suggest also to read the (a bit later) AntNet papers I wrote with Dorigo. I have a web page on the topic (nothing much, but there are some good references there):

      http://www.idsia.ch/~gianni/antnet.html

  6. Obligatory... by Totenglocke · · Score: 4, Funny

    How do you incorporate Swarm intelligence into computer AI? Simple - SPAWN MORE OVERLORDS!

    --
    "The tree of liberty must be refreshed from time to time with the blood of patriots and tyrants." ~Thomas Jefferson
  7. Re:Swarm UDP? by Wraithlyn · · Score: 3, Insightful

    I'm not sure I understand your post.

    Swarm logic would [...] prevent protocols like UDP from ever getting packets through in any sort of decent order

    UDP packet order is already unreliable. You have to build in your own sequencing and error-correction logic at the application level (if you need it).

    anything with ordered sequential data streams would see little to no benefit

    Again, UDP does not fit this description. Are you sure you're not confusing TCP and UDP?

    http://en.wikipedia.org/wiki/User_Datagram_Protocol#Comparison_of_UDP_and_TCP

    --
    "Mind, as manifested by the capacity to make choices, is to some extent present in every electron." -Freeman Dyson
  8. Console message by jewishbaconzombies · · Score: 3, Funny

    >format /dev/fd0

    >ANT: We must save the queen!
    >ANT: Which one of us is the queen?
    >ANT: I'm the queen!
    >ANT: No, I'm the queen!
    (smashing of glass sound)
    >ANT: Freedom! Horrible, horrible freedom!

  9. Re:And this is different from OSPF how? by Anonymous Coward · · Score: 3, Interesting

    Well, for one, ants don't do OSPF...

    It's been many years since I read up on this, but here goes: Like pheromone, the traces decay, or "evaporate", over time. Yet the next packet deposits new pheromone. Do this a number of times over a number of paths through a number of nodes, and statistically the shortest path will accumulate the most pheromone, and hence the shortest path will "emerge" from the rest.

    OSPF and hop counts are deterministic. This method is a lot more stochastic, and hence could be easier to scale it better to larger networks and problems like the TSP.

    Another advantage is that each node stores simpler information (the pheromone level for neighbor-destination pairs) instead of the whole network topology. Further, if a node goes down, the next best path is also readily known (better reaction to failures was one of their selling points). Also, it's possible (but probably not advisable) to start sending data packets without any kind of initial routing setup, and let the ACO thing figure out the best path over time. A lot of these points are very useful for, say, ad hoc networks, even if they may not make much sense for the Internet.

    It's been around for a while. Read up on it, it's pretty interesting... maybe starting with TFA.