· 653 days ago ·
Netflix Prize
$1,000,000 to the person who can improve the accuracy of collaborative filtering results over and above the results predicted by Netflix’s own rating system. The training data provided is a 10% subset of the total data collected by Netflix from past user film ratings.
10% is a high target but the competition runs until 2011 and the data itself is a rare example of a substantial real world problem so it should be interesting to work with.
· 905 days ago ·
Artificial Intelligence: A Modern Approach
One of the undisputed classics of AI literature is Artificial Intelligence: A Modern Approach by Stuart J. Russell and Peter Norvig. For anyone looking to get a comprehensive introduction to this exciting area, I would highly recommend this book.
The book’s table of contents, related online discussion of the book and the topics contained therein and many links to useful AI resources around the web are available on the pages Artificial Intelligence: A Modern Approach on Berkeley’s computer science website.
· 905 days ago ·
Python Robotics
When people think of Artificial Intelligence, they usually think of robots and/or films like Terminator. If this is your vision of artificial intelligence and you’d like to explore the area more, then Python Robotics is the place to start. “The goal of the project is to provide a programming environment for easily exploring advanced topics in artificial intelligence and robotics without having to worry about the low-level details of the underlying hardware”.
· 945 days ago ·
Machine Learning weblogs
Here’s two links seen on Jefferson Provost’s weblog to other machine learning weblogs. Unlike backprop.net which was always meant to be a link log (primarily for my benefit and if others find it useful then that’s a bonus), Jefferson’s site and even more so these two links that he posted strive to post original machine learning content:
· 985 days ago ·
Predictive Uncertainty Challenge
Another fascinating challenge: The Predictive Uncertainty Challenge involves the prediction of regression outputs to one synthetic and three real datasets. The predictions should be accompanied by an estimate of the certainty in a solution. From the challenge site: “The models will be ranked according to the average negative log-likelihood on unseen data, this will penalise inaccurate over-confident predictions as well as under-confident, but accurate predctions. Initially the models will be ranked according to the performance on the validation set, and performance on test data will only be revealed after the competition deadline.”
backprop.net -- All pages and content are Copyright © 2002-4 Robert Wall