Speaker-independent isolated digit recognition: Multilayer perceptrons vs. Dynamic time warping

By L. Eon Bottou, F. Fogelman-Soulié, P. Blanchet, J. Liénard. 1990

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling (2018, methodology): “They were used prominently for speech recognition in the 80s and 90s (Waibel et al., 1989; Bottou et al., 1990).”
Convolutional Sequence Modeling Revisited (2018, background): “Waibel et al. (1989) and Bottou et al. (1990) studied the usage of time-delay networks (TDNNs) for sequences, one of the earliest local-connection-based networks in this domain.”
Markovian Models for Sequential Data (2004, methodology): “In some cases [95, 92, 2, 93, 94], the ANN outputs are not interpreted as probabilities, but are rather used as scores and generally combined with a dynamic programming algorithm akin to the”
The general inefficiency of batch training for gradient descent learning (2003, background): “Several researchers clearly state that on-line training is faster that batch training, especially in pattern recognition problems with large training sets (Bengio, 1991, 1996; Bouttou et al., 1990).”
Gradient-based learning applied to document recognition (1998, methodology): “TDNNs have been used in phoneme recognition (without sub-sampling) [40], 41], spoken word recognition (with sub-sampling) [42], 43], on-line recognition of isolated handwritten characters [44], and signature veri(cid:12)cation [45].”