Paper accepted at ICDE 2020

We are proud to announce that our paper "Hierarchical Quick Shift Guided Recurrent Clustering" has been accepted at the ICDE conference (ranked A*)

We are proud to announce that our paper "Hierarchical Quick Shift Guided Recurrent Clustering" by Can Altinigneli (University of Vienna & Ludwig-Maximilians-Universität), Lukas Miklautz (University of Vienna), Christian Böhm (Ludwig-Maximilians-Universität), and Claudia Plant (University of Vienna) has been accepted at the valuable ICDE conference (ranked A*).

We present a framework in which our novel Hierarchical Quick Shift (HQuickShift) clustering algorithm supervises the training of a Recurrent Neural Network (RNN) to learn the underlying dynamics of the mode-seeking clustering process. Our algorithm supports variable density clusters with arbitrary shapes. It does not require the expected number of clusters and the difficult to tune neighborhood parameter, which controls under- and over-fragmentation of the modes. We introduce an intuitive parameter "min_mode_size" to select the smallest size grouping that we wish to consider a "mode-level-set". This trades a counter-intuitive parameter for one that is easier to select.

Check-out our implementation (anonymous review version) under http://bit.ly/2YTGBU8 which includes the datasets, the trained RNN models, the jupyter notebook to reproduce all figures and our experiment results with corresponding explanations related to software and hardware setup.