Distributed clustering of categorical data using the information bottleneck framework

Highlights：

• We implement existing Information Bottleneck based clustering algorithms and explore their limitations on a single machine, with large inputs.

• We propose two new implementations that manage larger datasets in multiple machine setups without compromising the overall clustering quality.

• We provide theoretical foundations for the proposed implementations based on representative sampling and an efficient merging strategy.

• We evaluate the performance of these implementations both with real and synthetic datasets.