Guided Stable Dynamic Projections
Projections attempt to convey the relationships and similarity of data points from a high dimensional dataset into a lower- dimensional representation. Most projections techniques are designed for static data. When used for time-dependent data, they usually fail to create a stable and suitable low dimensional representation. We propose two new dynamic projection methods (PCD-tSNE and LD-tSNE) based on the idea of using global guides to steer projection points. This avoids unstable movement that hinders the ability to reason about high dimensional dynamics while keeping t-SNE’s neighborhood preservation ability. PCD-tSNE scores a good balance between stability, neighborhood preservation, and distance preservation, while LD-tSNE allows us to create stable and customizable projections. We demonstrate our methods by comparing them to 11 other techniques using quality metrics and datasets provided by a recent benchmark for dynamic projections.
Use the menu on the left to have access to the data/code and explore the results.