Would be interesting if console games included smartphone apps that had more game content to augment the game on the console. Like mini-games to get more gold/money/experience/characters/weapons etc to use in the console game. Also, by playing the real game different smartphone minigames could be played.
To make money some of the games could offer in app purchases.
Watch Dogs did this but all you did was play against others that were playing it on the pc.
http://en.wikipedia.org/wiki/Ellsberg_paradox is a decision theory paradox. One solution to the paradox is to use this Choquet integral. I don’t have time to read about it right now, but it seems like something that would be interesting to observe in multi-agent learning algorithms. What decision do they learn to take, what would make them learn the Choquet integral method? Is it stable, meaning do they always chose a particular action in line with a particular method or not?
A paper on the subject: http://ces.univ-paris1.fr/membre/cohen/pdf/gradef.pdf
Seems like it is useful in multi-objective combinatorial optimization: http://eric.univ-lyon2.fr/~arolland/publis/LustRollandMIWAI2013.pdf
Maybe I was asking the wrong questions as no one is looking at it from a MAL perspective.
So, I think that I will need to augment how we make features, behaviors and targets in order to get the automatic feature selection (and now automatic target selection as well. I forgot about how features can take targets!).
We need to annotate them. Otherwise, I think that we won’t be able to get a good clustering. Also, annotation will help to pick the features and the corresponding targets. I think that maybe that induction from the targets for the behaviors to the features might be something to think about. The annotations are essentially providing the semantics that we as humans take for granted. Otherwise the clustering would probably not work.
I also think that targets will need to be chosen based on the overall semantics of the learned behavior. So, like the behavior for determining whether to play on offense or defense was based on a feature distanceTo(me, centerField) I would have to say that this behavior is semantically a location based behavior. This would then limit the set of behaviors and targets the feature and target selector would have to choose from. I think that this is enough. In the future this could be even more generalized.
If this method doesn’t work to pick the targets for the features, then the targets could be chosen in post by the trainer and then the decision tree could be built based off of that.
I think that fractal clustering could be used to cluster data streams of electricity data.
What about distributed clustering algorithms? So, we have massive data streams from different parts of the grid coming in to their respective hubs. How do you make sense of the whole? I think fractal clustering would work here. I believe that the data would be self-similar enough to warrant such an algorithm.
Using the Fractal Dimension to Cluster Datasets
Fast Feature Selection using Fractal Dimension (another interesting paper). This one made think that maybe it would be possible for Horde to automatically select the correct features to be using to make its decisions on. This would be extremely useful, however highly unlikely that it would work with the few examples it is currently classifying on. I think that in order for automatic feature selection it would need a live stream of the features and “spike” the features on transitions between behaviors. I would then be able to continue doing horde as normal but would base my decision tree off of the new feature set.
Would this also be useful in creating the hierarchies? I think so. I will have to think a bit more about this aspect.
Ohh how I wish I had read through my posts before going to Robocup. I had the chance to talk with Peter Stone!!! I could have discussed some of my ideas with him and maybe coauthored a paper with him on this. Oh well maybe I can do it at GMU.
Well, I just remembered I wrote a couple months ago about this topic I have been talking about. https://drewsblog.herokuapp.com/?p=1471 (Dynamic lane reversal and Braess’ Paradox). The paper referenced there even has a nice name for what I have been talking about.
WOW…
So, at least I finally remembered. I also took a more multiagent perspective to the problem. I am now intrigued to find out if my method would scale compared to their ILP approach.
I also found the first authors notes! http://www.cs.utexas.edu/~mhauskn/projects/aim/notebook.html. They are very detailed and even mention the Braess’ Paradox! He apparently encountered it. It seems like the project just stopped. Many interesting questions are still open regarding the problem. It is certainly not solved. Also, in there experiments they never had a road fully become one-directional there was always at least one lane that went in the other direction. The other thing was that lane was always on the side that it “should” be on. That lane never could change sides or be in the middle or something really strange. Which would be possible in my system.
It seems like bounties might be able to work in this environment. I have to think about it a bit. Peter has been doing more recent work on the subject and using Auctions to allow drivers to bid on things. He has also created larger scale simulations with a nice simulator for Austin Texas. http://www.cs.utexas.edu/~mhauskn/publications.html
Cool RL reading group at UT https://www.cs.utexas.edu/~eladlieb/RLRG.html
So, abstractly what is a road. It is a queue for vehicles. Intersections allow elements (vehicles) to change queues. In this thought experiment I will abstract away lanes to just increased bandwidth and a corresponding increased throughput. Essentially, the idea with dynamically changing road systems is two fold: one, we want the transitions between queues to be as efficient as possible two, we want the throughput of the queues to be maximized. Main cause of congestion is crossing traffic which holds up the queue. So, usually we create a separate lane for those that are turning against opposing traffic. Or we make roads one way and you can’t turn against traffic (seen in DC).
So, the problem turns into a resource allocation problem at intersections, meaning you have some number of queues (resources) and you have to get the elements (cars) as quickly through to there destination. The idea then would be to determine how to adjust the queues to increase the resources when necessary and at the right locations.
This is sort of random… Have you ever seen a Jacobs Ladder? I was thinking of putting a floating farm on a Jacobs ladder type structure.
http://edwinsanimatedimages.com/jacobs_ladder.htm
I’ll have to think later if how bad it would work…
Is my idea for the dynamic road systems scalable to current road networks or will new road constructs need to be developed? I think the current networks will work perfectly. Because the main construct I think will be the figure-eight. This will allow the system to change. This is already present in most networks in terms of blocks and circles.
If it works it would be incredibly useful, here are some quotes on the subject:
The World Bank estimates traffic congestion costs Egypt 8 Billion dollars per year, which is approximately 4% of Egypt’s GDP [1].
http://d2dtl5nnlpfr0r.cloudfront.net/tti.tamu.edu/documents/mobility-report-2012.pdf
[1] World Bank, 2012, “Traffic Congestion in Cairo: An overview of the causes as well as possible solutions,” Report 71852, Washington D.C.
When we have autonomous cars will we be able to have the ability to make roads bi-directional? Meaning during parts of the day one or both side of the road could be going in the same direction. Or sometimes switching (like going from USA style to driving like people in England). Or will our roads be constructed to not need to do this? Or is it actually less efficient to do that sort of thing and just keep everyone driving on the right side of the road?
My hypothesis is that given local communication and cooperation of the cars, a dynamic road rules system that can adapt to the needs of the traffic will be much more efficient than the static rules that we currently adhere to currently. Already we know of the crazy taxi drivers that can get you to your destination faster than anyone else and its because they are willing to break the rules. Of course this is extremely dangerous now, but when the cars agree to change the rules I think that it will be even safer than if they would continue to follow the current set of rules. For reasons just like the taxi driver that is willing to break the rules. So, I’m pretty sure patterns will emerge and norms will develop and an equilibrium established where at particular times of day the roads are like x and at another time they are like y. I believe that a system like this would also be better for emergency situations. The ability to dynamically adjust the traffic patterns along a route would be life saving. Also, logistics/trucking companies could purchase “routes” sort of like advertisements on google. Essentially giving their destination a higher priority than normal.
Knot theory? Seems like loops may develop and the cars might end up colliding or creating longer paths in order to avoid colliding.
Control theory and DA? Much of this would build on a reliable lane changing and driving control constructs. Hopefully a bit of distributed AI in there as well.
MAS/L? For the higher level decision making, cooperation, and goal setting, definitely relevant. The cars have to on their own form coalitions and make decisions that will most certainly produce ripple effects that they will not be able to predict but other coalitions must adapt to. These systems will be very large and complex. However, I believe that given that they will learn quickly (lots of info) and change at a reasonable pace that a MAS is called for here.
Is it homogenous. No! There are all types of vehicles and types of traffic. This makes it very interesting. What weight do they bring to the system.
Is it hierarchical? Seems like different locations might represent parts of the hierarchy. At least in terms of how much of the system they can effect by making changes to the rules. The hierarchy in this case would be highly dynamic based on time of day. Maybe there is an underlying hierarchy that might form naturally when looking at the coalitions of traffic. I mean this is essentially what happened when we added lanes to the highways. Right? We created a hierarchy of types and goals of the drivers. I think that not only will there be natural hierarchies, but I also think that there may be the need to impose or provide engineered hierarchies. This would help the system to more quickly adapt and realize the importance of events.
I think that I mainly had this idea because as I was waiting in a line to check out a small kid asked her mom why they had to stand in this long line. She reminded me of the need to nurture the ability to see past the norms of society, such as having to wait in line, and produce creative solutions. The kid is right, we really shouldn’t need to have to wait in a long line to purchase items from an outlet. We have the technology to eliminate them. Most of the time we can avoid lines by purchasing things online. But, we still are in the habit of purchasing food things from local stores. So, there is still the line. Giant has installed in some of there stores a scanner so that you can scan you groceries as you add them to your cart. For more than a few items this becomes a hassle. Safeway has a program where you can order your groceries and employees can deliver them to you. This does not seem efficient, or cost effective when only a few are doing this.
I dream of when you could walk in and your phone already has created a list of things that you probably will want and compiled the available coupons and deals for those items. You would then scroll through the list and approve and remove the selections. As you are browsing and finding the other things that were not on the list your order is being packaged by robots and humans in the back. The more standard your order is you could get reduced price.
Well that was a lot.
So there would be a variety of robots. There would be lots of mini scouter robots. These types of robots would be distributed throughout the grid. Their job would be to detect when a break occurs in the wire and then to determine the exact location. Then they would alert their commander and then either provide safety fixes, like cutting and capping wires or just setting up safety perimeters so no one gets hurt. Then with the information gathered by the scouts (pictures/video, electrical data, course taken ie cutting or capping wires, approximate number customers affected) the commander can determine which robots to send out to fix the problem. The decision to send out the type of robot will be done by a virtual agent. However, a human will receive the info as well in order to determine whether emergency teams should be deployed to rescue any injured people.