Yaser Sulaiman’s Blog

In addition to being a nuisance, spam (junk) emails waste user time, disk space and network bandwidth. On my way back to KFUPM after the summer vacation, a simple idea regarding spam filtering hit me.

It all began with a simple question: “Why don’t you want to see spam emails?“. The answer was straightforward: “Because I’m not interested in whatever subjects the email is talking about“. Then, I started thinking:

We can’t use the subject header of the email because it can be totally unrelated to the body. So, we have to look at the content itself. How about if we extracted keywords that represent the main subject(s) of the email, and then compared them with keywords that represent the subjects that the user is interested in? After that, we should come up with a predicted “level of interest“. If it is too low, then the user will (most probably) not be interested in seeing this email (i.e. it’s spam).

This idea is now the core of my senior project. It will be a research-oriented, AI-related project. Regarding the first phase (keyword extraction), my teammate and I will most probably use some of the available services. We will focus our efforts on the second phase (keyword comparison). We have to figure out exactly how to do it and how to incorporate machine learning in it. We also might improve it by using Bayesian Belief Networks and/or Functional Network classifiers.

Spam filtering is one of the hot topics in the application of data mining and AI techniques. By working in this project, we hope that we can contribute to the ongoing research and develop an approach that will hopefully be taken as a basis for a new filtering technique or as an addition to existing ones.

What will be the result of a game of checkers with no mistakes made by either player? A team at the University of Alberta said they had “solved” checkers: “perfect play by both sides leads to a draw“.

The research published in the journal “Science” says that dozens of computers have been working on solving checkers for more than 18 years by playing out about 39 trillion possible positions!

Jonathan Schaeffer, the chairman of the department of computing science at the University of Alberta said:

Clearly, the world is not going to be revolutionized by this. The important thing is the approach. In the past, game-playing programs have used rules of thumb – which are right most of the time – to make decisions. What we’ve done is show that you can take non-trivial problems, very large problems, and you can do the same kind of reasoning with perfection. There is no error in the Chinook result. … Every decision point is 100%.

This proof is what is called a “weakly solved” result, because it calculates the result from an initial position – 10 pieces on the board – instead of from the beginning of the game. Could it be possible to produce a “strong solution” (calculating every position from the beginning of a game)? “Maybe”, said Schaeffer, “but there is not enough computer power available”.

This breakthrough could encourage others to solve the other games we encounter in life:

Artificial intelligence technology has been used to generate strong heuristic-based game-playing programs, such as DEEP BLUE for chess. Solving a game takes this to the next level, by replacing the heuristics with perfection.