The output distribution of the state is a context-dependent multi-space Gaussian distribution [36], and these are clustered into groups of related contexts using a decision tree in order to reduce the number of free parameters and allow the modeling of unseen contexts.
In other words, this structure cannot accurately predict the parameters of models of unseen contexts, given the very limited subset of contexts observed in the training data.
Section 3 proposes an approach which enables the conventional structure to model the overlapped contextual factors and thus improves the modeling performance of unseen contexts.
However, this 'divide-and-conquer' approach leads to data sparsity, with the consequence that it suffers from poor generalization, meaning that it is unable to accurately predict parameters for models of unseen contexts: the hard decision tree is a weak function approximator.
Once the decision tree is constructed, unseen contexts can be taken into account and leaves containing little or very similar data can be merged (for more information on how trees are built and used, please refer to [30]).
Therefore, a context clustering approach such as decision tree-based clustering has to be used to decide about unseen contexts [31, 45].
Our hypothesis is that by enabling each training sample to influence multiple sets of model parameters (and thus cover a larger portion of contextual space), generalization to unseen contexts would be improved.
Comprehensive understanding of what really happens during and after a disaster like Fukushima perhaps requires awareness of the "unseen" social dynamics and historical context as well as the immediate physical effects that GSV allows scholars to access remotely.
In its largest context, dharma refers to the unseen and limitless web of intelligence that sustains and supports both the universe as a whole and each individual creation (person, plant, animal, even mineral) within it.
THE UNSEEN, by Heather Graham.
O.K., the unseen voice said.
