MaxWiK - Machine Learning Method Based on Isolation Kernel Mean Embedding
Incorporates Approximate Bayesian Computation to get a
posterior distribution and to select a model optimal parameter
for an observation point. Additionally, the meta-sampling
heuristic algorithm is realized for parameter estimation, which
requires no model runs and is dimension-independent. A sampling
scheme is also presented that allows model runs and uses the
meta-sampling for point generation. A predictor is realized as
the meta-sampling for the model output. All the algorithms
leverage a machine learning method utilizing the maxima
weighted Isolation Kernel approach, or 'MaxWiK'. The method
involves transforming raw data to a Hilbert space (mapping) and
measuring the similarity between simulated points and the
maxima weighted Isolation Kernel mapping corresponding to the
observation point. Comprehensive details of the methodology can
be found in the papers Iurii Nagornov (2024)
<doi:10.1007/978-3-031-66431-1_16> and Iurii Nagornov (2023)
<doi:10.1007/978-3-031-29168-5_18>.
