Matroid at Strata + Hadoop World

Neural Networks can be a challenge to train on Spark because of the communication-heavy nature of evaluating dense networks. At Strata + Hadoop World 2015 in New York, Reza presented the matrix computations and hardware acceleration primitives that allow neural networks to be trained on Spark.

See more about the “Hard Core Data Science” session at Strata:

 

Thanks to Ben Lorica for organizing, and putting together a summary of the event.