Limits...
Deep Neural Networks with Multistate Activation Functions.

Cai C, Xu Y, Ke D, Su K - Comput Intell Neurosci (2015)

Bottom Line: Experimental results on the TIMIT corpus reveal that, on speech recognition tasks, DNNs with MSAFs perform better than the conventional DNNs, getting a relative improvement of 5.60% on phoneme error rates.Further experiments also reveal that mean-normalised SGD facilitates the training processes of DNNs with MSAFs, especially when being with large training sets.The models can also be directly trained without pretraining when the training set is sufficiently large, which results in a considerable relative improvement of 5.82% on word error rates.

View Article: PubMed Central - PubMed

Affiliation: School of Technology, Beijing Forestry University, No. 35 Qinghuadong Road, Haidian District, Beijing 100083, China.

ABSTRACT
We propose multistate activation functions (MSAFs) for deep neural networks (DNNs). These MSAFs are new kinds of activation functions which are capable of representing more than two states, including the N-order MSAFs and the symmetrical MSAF. DNNs with these MSAFs can be trained via conventional Stochastic Gradient Descent (SGD) as well as mean-normalised SGD. We also discuss how these MSAFs perform when used to resolve classification problems. Experimental results on the TIMIT corpus reveal that, on speech recognition tasks, DNNs with MSAFs perform better than the conventional DNNs, getting a relative improvement of 5.60% on phoneme error rates. Further experiments also reveal that mean-normalised SGD facilitates the training processes of DNNs with MSAFs, especially when being with large training sets. The models can also be directly trained without pretraining when the training set is sufficiently large, which results in a considerable relative improvement of 5.82% on word error rates.

No MeSH data available.


The curve of the logistic function. Its definition domain is (−∞, +∞) and its range is (0,1). This is a nonlinear function which has an exponential increase in the middle, separating state 0 and 1 apart from each other.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4581500&req=5

fig1: The curve of the logistic function. Its definition domain is (−∞, +∞) and its range is (0,1). This is a nonlinear function which has an exponential increase in the middle, separating state 0 and 1 apart from each other.

Mentions: Figure 1 is the curve of the logistic function. It is easy to note that most of changes occur in the middle part. As x increases, y increases. The increase is significant when x is close to 0. By contrast, y is almost stable at 0 when x is far less than 0 and stable at 1 when x is far more than 0. To classify patterns, x can be set to the features of the patterns and the classification results are 0 or 1. The curve can also be translated by adding constants to (1), namely,(2)y=11+e−x+x0+y0,where x0 and y0 are constants.


Deep Neural Networks with Multistate Activation Functions.

Cai C, Xu Y, Ke D, Su K - Comput Intell Neurosci (2015)

The curve of the logistic function. Its definition domain is (−∞, +∞) and its range is (0,1). This is a nonlinear function which has an exponential increase in the middle, separating state 0 and 1 apart from each other.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4581500&req=5

fig1: The curve of the logistic function. Its definition domain is (−∞, +∞) and its range is (0,1). This is a nonlinear function which has an exponential increase in the middle, separating state 0 and 1 apart from each other.
Mentions: Figure 1 is the curve of the logistic function. It is easy to note that most of changes occur in the middle part. As x increases, y increases. The increase is significant when x is close to 0. By contrast, y is almost stable at 0 when x is far less than 0 and stable at 1 when x is far more than 0. To classify patterns, x can be set to the features of the patterns and the classification results are 0 or 1. The curve can also be translated by adding constants to (1), namely,(2)y=11+e−x+x0+y0,where x0 and y0 are constants.

Bottom Line: Experimental results on the TIMIT corpus reveal that, on speech recognition tasks, DNNs with MSAFs perform better than the conventional DNNs, getting a relative improvement of 5.60% on phoneme error rates.Further experiments also reveal that mean-normalised SGD facilitates the training processes of DNNs with MSAFs, especially when being with large training sets.The models can also be directly trained without pretraining when the training set is sufficiently large, which results in a considerable relative improvement of 5.82% on word error rates.

View Article: PubMed Central - PubMed

Affiliation: School of Technology, Beijing Forestry University, No. 35 Qinghuadong Road, Haidian District, Beijing 100083, China.

ABSTRACT
We propose multistate activation functions (MSAFs) for deep neural networks (DNNs). These MSAFs are new kinds of activation functions which are capable of representing more than two states, including the N-order MSAFs and the symmetrical MSAF. DNNs with these MSAFs can be trained via conventional Stochastic Gradient Descent (SGD) as well as mean-normalised SGD. We also discuss how these MSAFs perform when used to resolve classification problems. Experimental results on the TIMIT corpus reveal that, on speech recognition tasks, DNNs with MSAFs perform better than the conventional DNNs, getting a relative improvement of 5.60% on phoneme error rates. Further experiments also reveal that mean-normalised SGD facilitates the training processes of DNNs with MSAFs, especially when being with large training sets. The models can also be directly trained without pretraining when the training set is sufficiently large, which results in a considerable relative improvement of 5.82% on word error rates.

No MeSH data available.