how to decide what to read

[Sam Joseph]

This is exactly the sort of thing that the NeuroGrid project is trying to
achieve.

http://www.neurogrid.net/

I won't claim that we have the perfect solution just yet, but we're
working towards a system where you can easily publish meta-data about
whatever documents you care to.  Over time the system learns who you go to
for which information, automatically adjusting future recommendations on
this basis.  e.g. I often read and download pages related to "Fly
Fishing"that were marked up by John, so my NeuroGrid node learns to go to
John first when I search for things related to "Fly Fishing".

NeuroGrid is completely open source, the code is all in SourceForge, which
means that it

1. is still under development
2 will take some time before it is a turn key solution

I believe that integrating trust and learning into this sort of
distributed system is essential and that ultimately we will all benefit
from the approach, whether NeuroGrid is the actual system we end up using
or not.

CHEERS> SAM

Geoffrey Hinton wrote:

> In the old days, there had to be a way to decide what to print because
> printing and circulation were bottlenecks.  But now that we have the
> web, the problem is clearly how to decide what to read.  It is very
> valuable to have the opinions of people you respect in helping you
> make this choice. For the last five years, I have been expecting
> someone to produce software that facilitates the following:
>
> On people's homepages, there is some convention for indicating a set
> of recommended papers which are specified by their URL's.  When I read
> a paper I think is really neat, I add its URL to my recommended list
> using the fancy software (which also allows me to make comments and
> ratings available).
>
> I also keep a file of the homepages of people who I trust (and how
> much I trust them) and the fancy software alerts me to new papers that
> several of them like.
>
> Its hard to manipulate the system because people own their own
> hompages and if they bow to pressure to recommend second-rate stuff
> written by their adviser, other people will stop relying on them.
>
> Obviously, there are many ways to elaborate and improve this basic
> idea and many potential problems that need to be ironed out.
> But I think it would be extremely useful to have.
> Somebody please write this software.
>
> Geoff Hinton