Abstract
Proposed herein is a method to map compressed data representations to smooth probability distributions by distributing each data point's weight across its k-nearest distribution bins using Gaussian decay, creating a natural exponential falloff where closer bins receive exponentially more weight than distant bins. Unlike traditional methods that either create spiky distributions (nearest-neighbor-only) or over-smooth local features (global kernel density estimation), a bounded Gaussian approach, as proposed herein, confines smoothing to local neighborhoods while maintaining smoothness within those regions. The algorithm includes circular distance support for temporal data (ensuring hour 23 is treated as 1 hour from hour 0, not 23 hours away) and graceful degradation for edge cases, making it robust for production security analytics systems.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kacewicz, Ania and Dagnall, Jud, "GAUSSIAN-WEIGHTED K-NEAREST DISTRIBUTION POINT MAPPING FOR SMOOTH PROBABILITY ESTIMATION", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10234