Members of The Ensemble have made major breakthroughs in the accuracy and in the performance of recommender algorithms.

Accuracy Single programs created by our members achieved these results.
		Craig Carmichael created multiple models that had a quiz RMSE of 0.87xx.
		7.64% improvement Nicholas Ampazis and George Tsagas from the Feeds2 team created multiple models that had quiz RMSEs of 0.87xx. Their best result was 0.8787.
	7.85% improvement Edward De Grijs included a number of novel improvements in a single model that obtained a quiz RMSE of 0.8767. His algorithm is not a time model, has low complexity, and it automatically finds the optimum parameters.
		7.65% improvement Jeff Howbert (team 'Howbert') created a time based matrix factorization algorithm that achieved a quiz RMSE of 0.8786.
		Using some new ways of capturing long-term and short-term time effects, Larry Ya Luo, a.k.a. Dace created a directly trained model that can achieve 0.877x quiz RMSE. With further post-processing, 0.875x can be achieved by a single predicator.
		8.24% improvement Aron Miller created what might be the best single algorithm yet discovered (with a quiz RMSE of 0.8730 or an 8.24% improvement over Cinematch's original score). Email aroneus at yahoo
Performance Single programs written by our members were able to make extremely accurate predictions in record setting time.
		Nicholas Ampazis and George Tsagas from the Feeds2 team implemented all of their SVD models (except SVD++ variants) so that they converge in under 5 minutes.
	7.08% accuracy improvement in 8 minutes More than 3,000% performance improvement Wojtek Kulik created an algorithm that produced a 0.8840 quiz RMSE solution in eight (8) minutes (including reading raw data and writing result file) on an AMD 64 X2 Dual Core @5200+ machine, without using parallelism of any kind (like CUDA, OpenMP, etc.). Other published models take from 4 to 8 hours to approach this level of accuracy. His SVD implementation converges in less than one minute.
		7,500% performance improvement for kNN Bo Yang (known as Newman ! on the leaderboard and forum) improved the performance of standard K Nearest Neighbor algorithms by 7,500%. He decreased the time needed to calculate 316 million relationships between objects from 2.5 hours to 2 minutes. He used a PC with Intel Core2 Quad CPU (4 cores) @ 2.4G Hz, 2 GB of RAM, running Window XP. Most code was written in C++, built by Visual Studio 2008.