Fig 1.
Deep-deep neural network language models.
Table 1.
Details of n-gram vocabularies from the MCI and AD-type dementia datasets.
Table 2.
Percentages of transcript files for training, test, and validation sets for the MCI and AD-type dementia datasets.
Table 3.
% error and perplexity on MCI held-out test set.
(h = Hidden layer size; Bz = Batch size).
Table 4.
% error and perplexity on AD-type dementia held-out test set (h = Hidden layer size; Bz = Batch size).
Fig 2.
% Error of the D2NNLMs vs. DNNLM on MCI dataset with smaller number of hidden layers.
Fig 3.
% Error of the D2NNLMs vs. DNNLM on AD-type dementia dataset with smaller number of hidden layers.
Fig 4.
Perplexity comparison between the D2NNLMs and DNNLM on the MCI dataset with smaller number of hidden layers.
Fig 5.
Perplexity comparison between the D2NNLMs and DNNLM on the AD-type dementia dataset with smaller number of hidden layers.
Table 5.
Performance comparison with the LPOCV AUC on the MCI dataset, N = 38.
Table 6.
Performance comparison with the LPOCV AUC on the AD-type dementia dataset, N = 198.