Gave my talk on branding for startups. Discussed the Three R’s of branding and some of the particular issues of note for a young company trying to get established in an Internet driven economy.

Then I went late to a talk on the Erlang programming language. It was standing room only so live-blogging wasn’t practical. Cool talk though. The big idea here: Erlang is massively scalable! The rest was over my head :)

Now I’m in a talk on Bayesian algorithms for filtering – two groups combined for this talk, one interested in Bayesian analysis the other in AI and cognition (with a futurist spin). Thought this would be a more philosophical discussion because of the AI, but the Bayesians have numbers on their side so the talk is getting into logic and algorithms. Spam filtering is a popular problem for applying the power of Bayesian. Basically by recognizing user behaviors and aggregating behaviors across users and then create probabilities for saving and for scrubbing any particular message. So Bayesian calculations get the probablities that score likelihood of scrub and likelihood of save. Then another algorithm has to look at the balance between the scores to determine the final save/scrub decision. The goal is to have a system that continues to learn over time to get better over time. Surprise issue – you don’t want the system to learn too fast! If it does the system can develop biases that might move you away from desirable result. Learning at the right pace allows the system to aggregate enough scores to have more relevant outcomes.

What does this have to do with branding? As I mentioned an hour or so ago, I’m indulging my nerdy roots and hanging out at Barcamp Atlanta. The technology is driving everything these days. And I believe in the long run these technologies will influence marketing and buying behaviors, just like the web has.

Moderating is discussing filtering large data sets – question now about qualifying market data as another use case. Bayesian is good at putting info into buckets. Not as good for mathematical evaluation.

Could you use Bayesian to create real estate recomendations? Start learning behavior for a home buyer? Could the home buyer train the system fast enough to make it useful infiltering a databse of 100000 homes? (questions from Alan Pinstein) The experts say yes, this is a good application for Bayesian approach.

Conversation is moving to relevance engines, but it is 10pm so time to change rooms.

### Like this:

Like Loading...

*Related*

**Tagged:** barcamp, barcampatl, barcampatl08, barcampatlanta

## Barcamp 2 – more from Friday night

October 18, 2008 § Leave a comment

Gave my talk on branding for startups. Discussed the Three R’s of branding and some of the particular issues of note for a young company trying to get established in an Internet driven economy.

Then I went late to a talk on the Erlang programming language. It was standing room only so live-blogging wasn’t practical. Cool talk though. The big idea here: Erlang is massively scalable! The rest was over my head :)

Now I’m in a talk on Bayesian algorithms for filtering – two groups combined for this talk, one interested in Bayesian analysis the other in AI and cognition (with a futurist spin). Thought this would be a more philosophical discussion because of the AI, but the Bayesians have numbers on their side so the talk is getting into logic and algorithms. Spam filtering is a popular problem for applying the power of Bayesian. Basically by recognizing user behaviors and aggregating behaviors across users and then create probabilities for saving and for scrubbing any particular message. So Bayesian calculations get the probablities that score likelihood of scrub and likelihood of save. Then another algorithm has to look at the balance between the scores to determine the final save/scrub decision. The goal is to have a system that continues to learn over time to get better over time. Surprise issue – you don’t want the system to learn too fast! If it does the system can develop biases that might move you away from desirable result. Learning at the right pace allows the system to aggregate enough scores to have more relevant outcomes.

What does this have to do with branding? As I mentioned an hour or so ago, I’m indulging my nerdy roots and hanging out at Barcamp Atlanta. The technology is driving everything these days. And I believe in the long run these technologies will influence marketing and buying behaviors, just like the web has.

Moderating is discussing filtering large data sets – question now about qualifying market data as another use case. Bayesian is good at putting info into buckets. Not as good for mathematical evaluation.

Could you use Bayesian to create real estate recomendations? Start learning behavior for a home buyer? Could the home buyer train the system fast enough to make it useful infiltering a databse of 100000 homes? (questions from Alan Pinstein) The experts say yes, this is a good application for Bayesian approach.

Conversation is moving to relevance engines, but it is 10pm so time to change rooms.

## Rate this:

## Share this:

## Like this:

RelatedTagged:barcamp, barcampatl, barcampatl08, barcampatlanta