Slashdot Mirror


Java Program Uses Neural Networks To Monitor Games

tr0p writes "Java developers have used the open source Neuroph neural network framework to monitor video game players while they play and then provide helpful situational awareness, such as audio queues when a power-up is ready or on-the-fly macros for combo attacks. The developers have published an article describing many of the technical details of their implementation. 'There are two different types of neural networks used by DotA AutoScript. The first type is a simple binary image classifier. It uses Neuroph's "Multi-Layer Perceptron" class to model a neural network with an input neurons layer, one hidden neurons layer, and an output neurons layer. Exposing an image to the input layer neurons causes the output layer neurons to produce the probability of a match for each of the images it has been trained to identify; one trained image per output neuron.'"

27 of 100 comments (clear)

  1. Can it.... by Barny · · Score: 2, Funny

    Ohhh, can it tell me when to move and shoot as well? Hey then interface it with the keyboard and mouse inputs and all my games can play themselves (like masturbation for computers).

    Then I can do other things while having fun playing games.

    --
    ...
    /me sighs
    1. Re:Can it.... by fractoid · · Score: 2, Funny

      And he'll call you 'dawg'? :P

      --
      Rampant carbon sequestration destroyed the Dinosaurs' tropical paradise. I'm here to help repair the damage.
    2. Re:Can it.... by Gastrobot · · Score: 2, Interesting

      I wrote a program in Java that used an artificial neural network to play Warning Forever. I took an evolutionary approach to training the networks because it was an unsupervised task. The program started with a pool of random networks, determined fitness by whatever criteria I used on a given run, and bred the networks together to make a new generation.

      My program had no capacity to play the game with a different interface than a human. It actually read the values of the pixels on the monitor, processed them through its network, and yielded keystrokes.

      It never got very good and eventually I got bored and moved on. I have some ideas for yielding better results that I want to try someday. Here's what happened:

      1) Playing on lives didn't work because eventually a network would destroy the boss' offensive capabilities and hide in a corner. The game would never progress.
      2) When I tried playing the game on a timer and ranking networks by what level they got to it ended up not moving at all and just shooting forward to destroy some bosses.
      3) When I tried ranking by time survived the network would again just destroy the boss' offensive capability and hide.

      1 is technically a perfect solution for the fitness criteria that I supplied. 2 and 3 are both examples of local minima where the networks found an early solution that dominated the competition (other networks, not the boss) and thus the gene pool.

  2. Huh. by Count+Fenring · · Score: 4, Informative

    Probably no one cares, but that's the wrong "queues" there. They mean "cues."

    1. Re:Huh. by julesh · · Score: 2, Funny

      Probably no one cares, but that's the wrong "queues" there. They mean "cues."

      I was wondering why I had trouble parsing that sentence, but didn't spot the reason. Thanks. :)

    2. Re:Huh. by jonaskoelker · · Score: 3, Funny

      Probably no one cares, but thats the wrong "queues" their. They mean "cues."

      Broke that for you.

  3. The '90-ties are over by Anonymous Coward · · Score: 2, Informative

    Why on earth are people still wasting their time on Neural Networks? Sure, they have a catchy name, but everything else about them sucks. Today we have much more robust methods available, e.g. Relevance Vector Machines, etc.

    1. Re:The '90-ties are over by Anonymous Coward · · Score: 3, Funny

      I prefer ties from the 1960s.

  4. Hilarious Overkill by phantomcircuit · · Score: 4, Insightful

    So they designed and wrote a neural network for the sole purpose of identifying a limited set of icons? Seriously?

    They could have done this using conventional methods that would be significantly faster. Me thinks someone was just doing this for entertainment.

    1. Re:Hilarious Overkill by Another,+completely · · Score: 2, Informative

      Me thinks someone was just doing this for entertainment.

      Almost certainly, especially since a complete success would just mean they can play video games slightly more efficiently.

      This toolkit worked for them, but does using a neural networking toolkit mean that what you produce is a neural network? It looks like the output neurons are doing image matching, and the hidden layer is identifying interesting candidates from a stream. In their environment, interesting candidates are any box that ticks from dim to bright (so they can spot the re-charged state when it gets fully lit).

      As described, it sounds more like a pipeline than a network. They use training data, rather than hard-coding target images, but it's not clear to me that the training feedback goes between neurons. It looks like you could do unit testing on the individual neurons, which doesn't describe neural networks as I understand them.

      I still think it's a neat tool that they made, but would some AI geek out there like to comment on calling it a neural network?

    2. Re:Hilarious Overkill by julesh · · Score: 4, Insightful

      multi resolution analysis perhaps? an example of this method is wavelet decomposition.

      Which is even more processor-intensive than a moderately sized neural net.

    3. Re:Hilarious Overkill by cerberusss · · Score: 2, Insightful

      So they designed and wrote a neural network for the sole purpose of identifying a limited set of icons? Seriously?

      They used a library. And it's just an algorithm.

      --
      8 of 13 people found this answer helpful. Did you?
    4. Re:Hilarious Overkill by mcvos · · Score: 4, Insightful

      IN JAVA. Speed was obviously not in the design criteria.

      The '90s are over. Java is now one of the fastest languages around.

    5. Re:Hilarious Overkill by grimJester · · Score: 2, Interesting

      My reaction as well. My first thought was the neural network was used to learn/predict user actions based on what's on the screen and give sound cues tied to the predicted action. For example, when some critter spawns in front of you, the player usually casts Fireball -> network learns that and shouts "Fireball!" every time a monster spawns.

    6. Re:Hilarious Overkill by molecular · · Score: 2, Funny

      The '90s are over. Java is now one of the fastest languages around.

      Around where?

      Everywhere, that's the cool thing about java, you know.

    7. Re:Hilarious Overkill by daid303 · · Score: 4, Insightful

      Except on everything not x86. The speed you currently see on desktop/server java is only accomplished by very good JustInTime compilers. Which are tweaked for x86. So everything else runs java like crap.

      http://en.wikipedia.org/wiki/ARM_architecture And that includes a few bilion ARM prosessors used in mobile phones. Sure they can run java, but it's nowhere near as fast as C.

      Java is nice, but the 'runs everywhere' feature is the least interesting one of them all. I can run an emulated a full blown x86 on a 8bit microcontroller, but that does not make it useful.

    8. Re:Hilarious Overkill by WankersRevenge · · Score: 2, Insightful

      I didn't it was slashdot's five minute hate already. Jeez its early. Anyone have an stray pictures of Gosling that I can yell at? Being somewhat serious, why does the slashdot groupthink give C# a free pass whereas java gets all the hate? I haven't looked, but I assumed both are similar in performance. Is it because one of them is integrated right into gnome? Or is it because Java is the most popular language for enterprise development?

    9. Re:Hilarious Overkill by mcvos · · Score: 4, Insightful

      Is your application CPU-limited?

      No, it's developer-limited. For most applications, development time is a bigger issue than execution speed. Only for very heavily used low-level routines (OS stuff, graphics libraries, VMs, etc) is it really worthwhile spending extra effort on extreme optimisation.

      If so, is it *the* fastest language?

      I don't have any recent benchmarks, but I remember that back in the days of the Java 5 JVM, Java is about 10% slower than equivalent C++, which is pretty good. But since then, JVMs have gotten quite a bit faster. It would surprise me if Java was not on at least equal terms with C++ now, alhough highly optimised low-level C is still going to be faster. But that's also extremely tedious to code.

      Those are the questions one should be asking when picking a programming language.

      No, the main question you need to ask when picking a language is if your code is going to be maintainable, and how expensive you can afford your maintainance to be. That's still the main timesink in development.

      If your application is limited by the CPU, only the fastest language, C, will do for some routines. You may even consider using assembly or machine-optimized code such as Atlas

      You accidentally hit the nail right on the head there: C is not necessarily the fastest language, highly optimised custom assembly is. And any language is only as fast as it can be if the programmer knows what he's doing. Some language do more for you to make optimal code easy to write than others.

      Java development, in my experience, is more laborious than Python or Ruby. Unless you have big teams of developers who must work close together, I wouldn't recommend Java for anything.

      Oh, I agree, Java stopped being an easy development language quite some time ago, and moved to the side of the fast execution languages. This is also why I switched from Java to Ruby. However, I just might switch to Scala because recent JVMs are so incredibly cool. The power of Java these days is more in the awesomeness of the JVM than in the language itself.

      Even so, there is an enormous amount of support for Java. It is by far the biggest language for enterprisey server stuff. I think there are as many webframeworks for Java as there are for all other programming languages put together. This is one of the big stengths of Java, but at the same time, this architectural overload is also one of the major hurdles for starting in Java.

      However, my point was that Java is pretty fast, which it is. If speed is an issue, Java can be an excellent choice (unlike Ruby, for example). If speed is the only thing that matters, then highly optimised C or assembly is really the only option.

    10. Re:Hilarious Overkill by ClosedSource · · Score: 2, Interesting

      The point is that you have no way of determining what training data is "correct" because you don't know anything about what the NN is "looking at".

      There also no guarantee that the network will ever converge and if it does there's no way to know if it has converged to a local minimum which isn't the solution rather than a global minimum.

  5. Neuroph look pretty cool by physburn · · Score: 3, Informative
    As a programmer, Neuroph looks like pretty good implement of Neural Networks, it handers most of the network types, Got built in graphic view of the output. Look nice, but which like as not, won't help, because you can rarely debug a neural network by looking at it.

    I've use neural network and genetic programming a few time, in work. Its completely different to normal programming. Instead of understand a problem completely, and write a structured solution to the task. You get a network and try and train it until its output matches what you think the output should be, no programming involved.

  6. Basshunter by Tokerat · · Score: 2, Funny

    I just got that obnoxious song out of my head, and now they want to put the whole game in there? No thanks!

    --
    CAn'T CompreHend SARcaSm?
    1. Re:Basshunter by jonaskoelker · · Score: 2, Funny

      Next, they'll let players play the game through an IRC bot named Anna...

  7. Re:How about NO image recognition? by happylight · · Score: 5, Informative

    It's an external program that doesn't have access to the game's internal data structures. Basically it's a bot without hooks to the game and makes decisions solely by looking at the screen.

  8. Re:How about NO image recognition? by Moraelin · · Score: 4, Insightful

    As someone who's been writing external trainers for games for years (though admittedly it was some years ago), I can assure you first hand that accessing a game's internal data structures is indeed very possible.

    And even if I couldn't find that boolean, I'd at least try to hook the point where it tries to draw that icon.

    The idea of using image recognition on the screen is so horribly inefficient a method... I suppose it could be used if absolutely nothing else works, but really that's about it.

    --
    A polar bear is a cartesian bear after a coordinate transform.
  9. Re:Cheating? by maxume · · Score: 2, Insightful

    Don't play against jackasses. Makes public servers a bit harder to deal with, but it is an easy solution otherwise.

    --
    Nerd rage is the funniest rage.
  10. SecuROM has... by zepo1a · · Score: 2, Funny

    SecuROM has detected JAVA on your computer! Please uninstall JAVA before launching the game.

  11. Re:How about NO image recognition? by skelterjohn · · Score: 2, Interesting

    Except to do something like that you have to analyze the program code of every game you make a trainer for. I've never done that sort of thing, but it seems scary.

    An approach like the one they've outlined can probably be moved from game to game with only parameter tweaks.

    That is the true beauty of machine learning :)