Sample Summary with Generative Encoding
share
With increasing sample sizes, all algorithms require longer run times that scales at best logarithmically. A concept that summarises the sample space to reduce the total number of samples into a core set that can be used for regression tasks is introduced. This idea of summarisation is called folding - the technique for projecting data into a lower dimensional subspace, whereas unfolding projects it back into the original space. Results for a prediction task show that information is retained during folding as accuracy after unfolding is still comparable to prediction without summarisation.
READ FULL TEXT