Eco-feller: Minimizing the Energy Consumption of Random Forest Algorithm by an Eco-pruning Strategy over MLC NVRAM
Hosted in Virtual Platform
DescriptionRandom forest has been widely used to classifying objects recently because of its efficiency and accuracy. Plus, non-volatile memory has been regarded as a candidate to replace DRAM. Random forest tends to construct lots of decision trees then conducts some post-pruning methods to remove the trees with less contribution. However, it would be a waste to write the to-be-pruned trees on non-volatile memory because of the high-cost write. This work proposed a framework to ease such waste by evaluating the importance of trees before constructing it then choose the proper modes to write data to non-volatile memory.