I just read an article about whale falls. They say the only way to find them so far is basically by accident. They describe these zombie worms that have been so far found at the sites of these falls. Why don’t they inject something into the worms that is traceable and transferable to its offspring. Then they could easily find the whale falls.
thoughts
Q-Learning with delayed updates
I’m sure someone has thought of this, but I didn’t look.
So, what if we delayed the update of the q-value and the policy for x number of time steps. We would keep track of the history and create an average reward. Initialize the policy for each action to be 1/|A| so that we can start with a large x. Then we can update q-value and the policies for each of the actions based on that average reward. Finally we can modify x based on a damped sine wave as time goes on.
This is similar to a lenient learner, but may also work against competitive players as well. The time where I average is sort of like a learning period to see what type of opponent and player I am playing against. This is what is like the lenient learner.
ScraperWiki
I found this cool service call ScraperWiki. It allows you to write python, ruby or php scraper as well as a view. You can make the data public so others can view or private. You can also make a view for your data! So I think this is awesome!!! I think for my Adv. AI class I think we could either use someone else’s data or make a scraper and mine our own data.
I also found Gravatar. Cool service that allows its users to upload an avatar so that whenever they post to a site that uses Gravatar, their avatar will be used! Really cool. The cool part is that you can get the avatar image based on a hash of the user’s email address! So maybe I could make an email address scraper and use that data to collect users Gravatar images and profiles! Really crazy all the info that users can put into their profile. Or I could maybe scrape for gravatar email hashes… Then directly get the user’s profile…
Not sure what I could do with that data though… Maybe try and find faces in the gravatar since their avatar may be their face. This could be useful if I can’t get access to their profile. Since then I could try and find them on facebook, twitter, etc..
I’ll have to get to you on what it would be good for… Sounds super creepy though that this could be done.
Bayesian game?
What if we have a stochastic game S and a set of agents A and in each game s we might have a different subset of A playing? What type of game is that? This of course is only relevant when dealing with |A| > 2 agents. Also, the agents that play in each s could be stochastic or fixed. Of course we can always make it harder by making the number actions varied for the agents.
This seems like it could model tournaments if the games are competitive or stochastic coalition games.
This seems sort of like a Bayesian game… Not sure though.
Stochastic Coalitional Game Theory
In my MAS class I am creating a stochastic normal-form game engine and one of the suggested elaborations is to look at coalitional games. So, I was like what if the coalition games were played in a stochastic game setting?
So what is that? Well based on these people at USC it is:
A real-world adversary is often a collection of distributed agents that must communicate and coordinate to enact a joint action and whose the [sic] motivations may not be perfectly aligned. This coordination in particular introduces new issues of stochasticity involving coalition structures, information uncertainty, imperfect execution and robustness.
So, it seems they took the location of the stochasticity further than just the games themselves, but to the robustness of the coalition structures as well. This is one of their technical approaches to their research. However, they seem to be the only ones looking at this type of game theory and they have not mentioned it in any of there papers that I can tell.
Swarm of text
Can we use the Shapley idea to determine if there is a swarm on twitter based on tweets? A swarm could be thought as a lot of retweets or a tweet with similar words. We could find old tweet swarms characterize them and use them to predict whether certain tweets will become swarms. I think the idea is identifying whether a tweet will go viral. When classifying we can look at the social network of the originators and the topic. Then we can see which words are most frequent in the emerging swarm. If pretweeting (stockmarket on the words in twitter) was still around we could make money :).
Is twitter and your (virality index) related to the success of crowd funded operations? Or just that you know some rich people.
How to bring about a virtual swarm and influence it in your favor is an interesting topic. Sounds like marketing though…
Swarms within the internet
I was wondering if I had access to the data that AddThis collects if I could identify swarming within the data using the Shapley value idea? Might be interesting.
No Free Lunch (NFL)
My friend Indranil introduced me to the theory of “No Free Lunch”. From Wikipedia article:
A conventional, but not entirely accurate, interpretation of the NFL results is that “a general-purpose universal optimization strategy is theoretically impossible, and the only way one strategy can outperform another is if it is specialized to the specific problem under consideration”
It is definitely something that I must keep in mind whenever doing research.
Another too difficult CV problem
Take a picture of equations and convert to text to be able to run through something like wolfram alpha.
Distributed 3d modeling using smartphones
You and your friends want to make a 3d model of something big!! Not just a toy or person. Rather your neighborhood or house. Something big. So the idea is to stitch together distributed 2d video to create a 3d model. You would use your smartphones that have gps, bluetooth, accelerometer, altitude data and compass. Send that data (of course would have to some sort of heuristic) back to a server which processes it and creates the model.
From researching this idea I found this article about how the navy developed malware for android to create a 3d model of the victims house. So the idea of creating 3d model from cellphone data is doable. I just want to distribute it and make it applicable to larger environments in real time.
You could become part of an initiative or start one. Essentially someone defines the boundaries of the model via gps coordinates. Then users who want to contribute data can join the network. Bluetooth could be used to enhance the quality of the mesh by providing another source of location data.
The mesh could then be imported into games like second life or maybe into google maps. Could use the 3d model to create the basis for an augmented reality platform. So if you tag something while you are viewing the 3d model on your computer and then you go to that place you could view it through your phone. The uses for a 3d model are endless.
Since we are distributing the data and the mesh will be updating, we could develop a highly robust to change mesh. Essentially free from dynamic objects. So if in one frame there is like a bike and in another there isn’t then I could determine to pick the frame without the bike.
This idea is really starting to sound like a huge project. For my advanced AI class I think I need to focus on one thing. Like the distributed 2d->3d modeling. This post became a lot longer than my initial idea.
This blog looks interesting.