June 4, 2004
Model Details Paper Released
|
After a meeting yesterday with Henry, I think that we convinced ourselves that we have the details of the activity inference model worked out. For the benefit of the hordes of people who would like to know the structure of the DBN that we are proposing, how an activity graph is compiled to a DBN and the conditional probability tables of that DBN, I present the following file. |
June 1, 2004
Automatic Web News Extraction Using Tree Edit Distance
|
This is a paper that was in WWW2004. It's goal was to analyze a bunch of Brazilian news sites and pull out bits of the news story from them. Things like the title, the author and the content. The approach that they took was to crawl a bunch of news pages, then cluster them according to their template, generalize the template, then analyze the portions of the templates that were changeable for the things that they wanted. The key component of the research and the reason why I was interested was because they have a nice overview of work in tree-edit distances. This is the metric that compares two tree data structures and assigns a similiarity score to them. |
I think that something like this is going to become important to the work with the anesthesiologists because we are going to have to generalize the data stream that we receive from them in some way. I'm anticipating that we will have some structure for the activity provided, but have to generalize the rest and the tree approach seems pretty good.
The one caveat to using this technique is that I don't think that it is going to work very tightly with a probabilistic model because the basic assumption in the web sphere is that if you see something then it is intended to be there. In our case if we see something it might not have been intentional and if you don't see something, maybe you should have.
May 31, 2004
Post-Mortem on RFID Activity Inference UBICOMP Paper Reviews
|
The following is my summary of the paper reviews that we received for the RFID Activity Inference Paper submission to UBICOMP (that was, unfortunately, rejected). The biggest issue that is recurrently coming up is why we need so much machinery to recognize activities. Wouldn't a simple object to activity mapping be sufficient. This is a valid and well-trodden comment. I am working on addressing these issues currently. More details about model constructions were requested. This would follow naturally from a discussion of why simpler models fail. (If in fact they do.) What is RFID tag detuning? This was mentioned, but I have no idea what it is referring to. |
I need to review Horvitz's work on activity recognition to understand how he used partial orders and timing. That work was mentioned (probably by him!), but I'm not sure that I understand how that info was incorporated in Lumiere.
Everyone can find something to critique about the data collection, but I think that if I can sufficiently justify the models then the data collection will be good enough.