Web Analytics

Partitioned Decision Trees for Line-Rate In-Network ML

How Does SpliDT Work?

How Does SpliDT Perform?

*Pareto-Optimal Performance

SpliDT consistently dominates the Pareto frontier, delivering higher accuracy for the same number of flows in the data plane

Pareto

*Latency Without Compromise

SpliDT matches the baseline detection latency while pushing the accuracy frontier

TTD

Acknowledgement

Sponsors:
Purdue University, University of Michigan, NextGArch Laboratory


Home