Firas Abuzaid | Joseph Bradley |
Feynman Liang | Andrew Feng |
Lee Yang | Matei Zaharia |
Ameet Talwalkar |
Partitioning by column unlocks the potential for more optimizations:
# instances | 8.1 $\times$ 10^{6} |
# features | 784 |
Size | 18.2 GB (sparse) |
Task | classification |
# instances | 2 $\times$ 10^{6} |
# features | 3500 |
Size | 52.2 GB (dense) |
Task | regression |
# instances | 2 $\times$ 10^{6} |
# instances | 2 $\times$ 10^{6} |
Task | regression |
Dataset | MNIST 8M |
Depth | 10 |
Check out the full paper at NIPS 2016!
Any questions? Shoot me an email:
fabuzaid at cs dot stanford dot edu