Introduction
ICF-MI combines two statistical algorithms, TT-ICE which is an incremental tensor-train decomposition alogorithm suitable for data streams and GIM which is a model identification alogorithm. The Algorithm has the following salient features:
TT-ICE is a novel incremental tensor-train (TT) decomposition algorithm which is developed to decompose high dimensional data streams into tensor-train format.
TT-ICE uses orthogonal projections and sequential application of SVD to compute the missing orthogonal directions in the TT-cores. For a stream of d-dimensional tensors, TT-ICE accumulates the stream along an additional (d+1)-th dimension and trains d 3 dimensional TT-cores that span the first d dimensions of the stream. The (d+1)-th TT-core resulting from the decomposition process contains coefficient vectors unique to each datum. Within the scope of our pipeline, we will treat TT-ICE as a multilinear dimensionality reduction tool and use the coefficient vectors contained in the last TT-core to train the surrogate model.
GIM is an information-theoretic model identification algorithm which can be used to identify the model which best fits the data.
ICF-MI is a combination of TT-ICE and GIM and can therefore be used to identify the model parameter in a cost efficient manner.
REFERENCES:
Google , search engine has been used throughout the project.
Aksoy et al., An Incremental Tensor-Train Decomposition Algorithm, arXiv preprint
Other kind of text
Bold reference.Bold letters.