Numerai is a new kind of hedge fund that pays out a reward to data scientists that can improve their meta model. The idea is to take a large number of uncorrelated high performing models and ensemble them. Money is divvied out based on leaderboard ranking awarding the top 100.
I tried out an xgboost model a wide and deep model using tensorflow. Google has used this wide and deep model to recommend apps on the Google Play store. In this algorithm a wide linear model and a deep feed-forward neural network are jointly trained. This approach combines the strengths of memorization and generalization.
Wide and Deep Learning research paper:
https://arxiv.org/abs/1606.07792
The xgboost model was rather poor but the tensorflow model ended up scoring the same logloss as the top 10 performing models.