On Sun, 26 Apr 1998, Jesse Montrose wrote:
> I've had some interest in this area myself, I would enjoy hearing about your
> progress.
>
>
> http://wearables.www.media.mit.edu/projects/wearables/augmented-memory.html
> I assume you've looked at this one, right?
yes, I made sure to look at this before I asked any questions, but it
wasn't exactly what I was thinking of for a rememberance agent.
Yeah, the RA is really designed for showing text relevant to your current
situation (especially when you've got a lot of text / context available),
but doesn't work well at all for small bits of text like "when is my
birthday." It doesn't have any natural language processing -- it's all
statistically based (using Term Frequency / inverse Document Frequency).
> Neural Networks can be built to run just fine on a 486, and you might find
> the processing done in the savant portion of project above to be a good
> place to start if you want to take that route.
Savant (the back-end to the RA) doesn't use neural nets -- it's mainly a
TFiDF text-retrieval system (similar to SMART and a lot of the web search
engines), along with a template system to recognize different kinds of text
documents and parse out & search on extra meta-information like source of a
document, time/date/day, location, etc.
-- Brad
BTW, I'm not on wear-hard anymore, so if you've got questions or comments
make sure to cc: me. Thanks for the pointers on ThoughtTreasure --
interesting stuff.
From Wear-Hard Mailing list Archive (WH)
Maintained by R. Paul McCarty
Archive created with babymail