Continuous Matrix Approximation on Distributed Data

Provided by: Association for Computing Machinery
Topic: Big Data
Format: PDF
The authors provide the first protocols for monitoring weighted heavy hitters and matrices in a distributed stream. They are backed by theoretical bounds and large-scale experiments. Their results are based on important connections they establish between the two problems. This allows the user to build on existing results for distributed monitor of heavy hitters, and extend them to weighted heavy hitters and then matrix tracking. Interesting open problems include, but not limited to, extending their results to the sliding window model, and investigating distributed matrices that are column-wise distributed.

Find By Topic