Abstract | ||
---|---|---|
Unsupervised clustering takes human brains a split second to complete in 2D space, while existing algorithms require many iterations involving all data points and an initial number of clusters ‘k’ to provide meaningful results. This initial ‘k’ cannot be provided by human if the data is in higher dimension where visualization is practically impossible. Attempts to calculate this value have low performance and give ambiguous results which are unhelpful to human judgment. This presents great motivation to search for a method to provide that initial ‘k’ in higher dimension. Human brains naturally group things in proximity together. By imitating this process and creating a middle process of grouping the data into subregions and mapping the data in each region into a bitmap of data densities, estimating the centroid locations and number of clusters can be simplified into a process of local maxima detection. A run of the algorithm on 2D data proved that it was effective for data with Gaussian-like distribution with some tolerance to overlapping. The algorithm therefore has great potential for data of higher dimension after generalization. This algorithm gives unambiguous initial ‘k’ and fairly accurate estimation of centroids which can speed up various popular clustering algorithms, including the k-means and Gaussian mixture models. Future research on middle grouping processes in human cognition, which may prove valuable in providing better machine learning algorithms, are also called for. |
Year | Venue | Field |
---|---|---|
2018 | HCI | Data point,Pattern recognition,Computer science,Visualization,Maxima and minima,Artificial intelligence,Bitmap,Cluster analysis,Mixture model,Centroid,Speedup |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
5 | 7 |
Name | Order | Citations | PageRank |
---|---|---|---|
Chloe Chun-Wing Lo | 1 | 0 | 0.68 |
Markus Hollander | 2 | 0 | 1.35 |
Freda Wan | 3 | 0 | 0.68 |
Alexis-Walid Ahmed | 4 | 0 | 1.01 |
Nikki Bernobic | 5 | 0 | 0.34 |
Nick Nuon | 6 | 0 | 0.34 |
Michael Shrider | 7 | 0 | 0.68 |