Month: May 2005

Posted on

Blogging after a while

I haven’t written for a while in this blog – I know this is not the way of looking after a blog! There are a few reasons that I didn’t post anything there:
1-My monitor was damaged (its vertical sweeping voltage generator was corrupted) and I couldn’t connect to the Internet using my own computer. I don’t feel easy to post from our lab.
2-I have started writing my thesis and it takes times and more than it mind!
3-I wanted to post a really useful(!) post, but it takes time to prepare one.

Well! That’s all!

Posted on

Myths of Reinforcement Learning

RL is slow?
Large state spaces are hard for RL?!
RL does not work well with function approximation?!
Non-Markovianness invalidates standard RL methods?!
POMDPs are hard for RL to deal with?!
…

In the opinion of Satinder Singh, these are myths of RL. He answers to these myths and expresses that these are not true beliefs. Moreover, there are many other information about reinforcement learning (myths, success, and …) in The University of Michigan Reinforcement Learning Group.

Posted on

CFP, Job Opening, and Weblogs

I read this Machine Learning mailing list policy for submission that suddenly a thought came to my mind. Read the shortened policy first:

* Meeting announcements. ….
* Publications of books or journal issues. In general, we accept announcements about the publication of multiple papers (e.g., journal issues) or books relevant to machine learning. Announcements about single papers such as articles or dissertations are not usually included.
* Career opportunities. …
* Software releases. …

I wonder if one can announce the release of a new academic weblog to such lists?! I think academic weblogs can change the way we do research and connect with each other significantly. Reading IlliGAL weblog or John Langford’s Machine Learning or a few others is a more useful for me than nth national XYQSD conference that I cannot attend. So, why not introduce such weblogs in our mailing lists?! (Of course, I am against those people who send their homepage or weblog address everywhere irrelevantly.)

Posted on

Don’t Send them to Prison

Once upon a time, Najmeh Omidparvar (in Persian) and Mohammad Reza Nasab Abdolahi (in Persian) lived together happily. However, the day came that they wrote something wrong in their weblogs, so they must have been jailed. They went to prison to continue their life there – not supposedly to be happy anymore. They are sent to different cells to live alone and think about themselves. Well … they are supposed to be alone. BUT, they were not that alone. Their souls are together, AND they had their child. Yeap! They had their child in the prison! (Read More)

Posted on

Some Experiments with Memetic Algorithm

A week ago, I tried some simple idea of what is called Memetic Algorithm with my friend, Ali Reza. He is working on Vehicle Routing Problem (VRP) for his MS thesis (industrial engineering) using evolutionary techniques. I told him about MA and its benefits; he became eager about the approach; we find a computer and we implemented a local search to put in the [cultural] evolution cycle. My local search was something like this:

Suppose that we have a route specified by R: c(1)c(2)…c(n) in which c(i) is the index of the ith city (or delivery position). I suggested this simple search:

For every position in the sequence, swap two adjacent cities and calculate the fitness, i.e. R: …c(i-1)c(i)c(i+1)… —> R’: …c(i-1)c(i+1)c(i)… and comparing f(R’) and f(R). If this change was better, accept it as the new individual.
This is very simple and I dare to say it is really mine. I guess someone might discovered (or used) this heuristic in their combinatorial optimization problems. However, I am not aware of such a search. Is there any in the literature?

We did not test it very much on many different problems. We have tested it in a single problem two or three times and in all cases, we got a better results comparing simple genetic algorithm. How much better?! I don’t know what amount is significant in this problem, but we reached (~)8.5 score when simple GA reached 9.5-10.

All of these has been done in less than an hour!

Mmm … As I told before, I do not believe in this type of interpretation of meme. However, I have tested it too!