No job, but research
Antabuse Online Buy Erythromycin Zyban Online Buy Soma Prednisone Online Buy Lotrisone Lipitor Online Buy Lipitor Erythromycin Online Buy CoumadinUnfortunately, I wasn't invited to a second round of interviews to either [Microsoft->The first interview] or [Amazon->http://seattle.henko.net/2005/11/16/amazoncom-interview/]. Although that is unfortunate, I don't feel too sorry about it. There will be a new recruiting wave and a new set of interviews in the spring, so I'll make another attempt then.
On the other hand, I did get accepted to do [research on the KnowItAll project->Research?]. More precisely I'm going to work on a program called TextRunner, which is a search engine which does not return documents containing certain terms, but answers questions. It can for example answer questions such as which the highest mountain in Sweden is, or return a list of American scientists. At the moment, this search engine extracts information from paragraph text on web pages, and my work is to make it also extract information from lists and tables on web pages.
I am actually very excited about this for several reasons. First, because I think TextRunner is the most interesting project of the several projects I've referred to as KnowItAll so far. But also because I believe lists and tables on web pages can contribute with very high-quality information to the system, so my work might end up being a very important part of the project.
Yesterday, I attended my first weekly meeting with the group of researchers working on KnowItAll and related projects. It was actually surprisingly similar to a typical student group meeting at Chalmers. Today, I talked to one of the main developers about the work I'm supposed to do, and he also explained to me how the system works, how I can get access to the research cluster, and other things that I need in order to work on the project.
Pardon me for making a technical digression here, but speaking of the research cluster, it is actually pretty cool. TextRunner is running on a cluster of 20 dual-Xeon machines, each with a single local disk of 250GB, resulting in a total of 5 TB(!) of storage space used to store a set of more than 90 million web pages. After being processed by TextRunner, this results in a graph structure with more 227 million nodes representing nouns such as "Albert Einstein", "car", "orange", and a mind-boggling 1.9 billion edges representing relations between them such as "is a", "worked for", "died in". I guess it's kind of nerdy, but I think it is really cool. 
Show this on a map.
2 comments November 22nd, 2005