HTML
-
For in-depth discussions of technical details such as temporal difference learning, Monte Carlo learning, eligibility traces, and Q-values, we refer the reader to Sutton and Barto (1998).
-
Supervised Learning (SL) algorithms are machine-learning algorithms, which generate a function that maps inputs to desired outputs.
-
A CL is a number between 0 and 1 based on acoustic measurements and defines how sure the system is to have performed correct recognition.
-
Multivariate linear regression (see page 1433 of Sheskin (2007)) models numerical data by a least squares function which is a linear combination of the model parameters and depends on >1 independent variables. A least squares function fits a model so that the sum of the squared residuals has its least value, a residual being the difference between an observed value and the value given by the model.
-
R2, the ‘coefficient of determination’ (see page 1230 of Sheskin (2007)), is the proportion of variability in a data set that is accounted for by a statistical model. R2 = 1 indicates that the fitted model explains all variability, R2 = 0, no ‘linear’ relationship between the dependent and independent variables, and R2 = 0.39, that approximately 39% of the variation in the dependent variable can be explained by the independent variables, and the remaining 61% by unknown variables/inherent variability.
-
In an n-fold cross-validation, the data is first divided into n (usually equal-sized) portions, and then in each of n folds, a different one of these portions is used for testing, while the remainder of the data is used for training. Results are averaged across the n folds.
-
-
A speech recognizer may provide a list, in order, of its top n hypotheses for a user utterance according to their CLs.
-
This is obviously also a very important issue for how best to evaluate the accuracy of user simulations.
-
If i = 1, i − 1 is considered to be the final slot, and if i is the final slot, i+ 1 is considered to be the first slot, for example, ‘So you want to fly from Edinburgh to where?’.
-
Utterances that cannot be handled by the system.
-
If there are no unfilled/unconfirmed slots to switch focus to, the strategy continues to follow DA2.
-
Copyright © Cambridge University Press 2009 2009 Cambridge University Press
Matthew Frampton, Oliver Lemon. 2009. Recent research advances in Reinforcement Learning in Spoken Dialogue Systems. The Knowledge Engineering Review. 24:166 doi: 10.1017/S0269888909990166
|