Improving the Effectiveness and Efficiency of Stochastic Neighbour Embedding with Isolation Kernel

This paper presents a new insight into improving the performance of Stochastic Neighbour Embedding (t-SNE) by using Isolation kernel instead Gaussian kernel. outperforms in two aspects. First, use t-SNE overcomes drawback misrepresenting some structures data, which often occurs when is applied t-SNE. because determines each local bandwidth based on one point only, while derived directly from data space partitioning. Second, yields more efficient similarity computation data-dependent has only parameter that needs to be tuned. In contrast, data-independent increases computational cost determining n bandwidths for dataset points. As root cause these deficiencies kernel, we show simply replacing with significantly improves quality final visualisation output (without creating misrepresented structures) and removes key obstacle prevents processing large datasets. Moreover, enables deal large-scale datasets less runtime without trading off accuracy, unlike existing methods speeding up