I cannot stress enough my pleasure at being invited to join this wonderful team and participate in such a unique experience.

I am not quite done with the Netflix Prize yet, as I’d like to describe here in particular an excellent model that I was fortunate to discover that can get a 8.24%+ improvement over Cinematch. At the time of writing this, I have not seen a claim or published model that is better. However, as we know, such things do not last long. This is progress!

I believe the most similar model in existence that has been published is Yehuda Koren’s and he refers to it as the TimeSVD++. His model in turn owes many references to innovations before his. I owe him and many other participants a thanks for not only making these achievements, but sharing them with the world, so that people such as I can hack improvements onto them.

One of the joys I hope to get out of publishing this model, is to have people immediately notice the things that I missed that can make it even more powerful. The power of multiple minds working on a problem has never been more clear to me then after the experience I just had with The Ensemble.

I will be checking email sporadically and will be back in-town by August 1st.

- Aron Miller

Original On Teh Road.