Meta-Sketch: A Neural Data Structure for Estimating Item Frequencies of Data Streams

To estimate item frequencies of data streams with limited space, sketches are widely used in real applications, including real-time web analytics, network monitoring, and self-driving. Sketches can be viewed as a model which maps the identifier stream to corresponding frequency domain. Starting from premise, we envision neural structure, term meta-sketch, go beyond basic structure conventional sketches. The meta-sketch learns sketching abilities meta-tasks constituted synthetic datasets following Zipf distributions pre-training phase, fast adapted (skewed) adaption phase. Extensive experiments demonstrate performance gains offer insights into our proposals.