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.


Results of classifying random points by using the neural network based on the binary XOR. The points are separated into 4 classes and 7 districts.
© Copyright Policy
Related In: Results  -  Collection

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

fig8: Results of classifying random points by using the neural network based on the binary XOR. The points are separated into 4 classes and 7 districts.

Mentions: The neural network can also classify the other points, as illustrated in Figure 8. The coordinate plane is separated into 4 classes and 7 districts approximately. A glance at the figure reveals that these districts are symmetrical for the diagonal of the plane. Although the numbers of the points are different in terms of the classes, the widths are almost the same. More importantly, the results are quite different from those in Figure 7, as the points marked by the same number can be in two districts which are far away from each other.


Deep Neural Networks with Multistate Activation Functions.

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

Results of classifying random points by using the neural network based on the binary XOR. The points are separated into 4 classes and 7 districts.
© Copyright Policy
Related In: Results  -  Collection

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

fig8: Results of classifying random points by using the neural network based on the binary XOR. The points are separated into 4 classes and 7 districts.
Mentions: The neural network can also classify the other points, as illustrated in Figure 8. The coordinate plane is separated into 4 classes and 7 districts approximately. A glance at the figure reveals that these districts are symmetrical for the diagonal of the plane. Although the numbers of the points are different in terms of the classes, the widths are almost the same. More importantly, the results are quite different from those in Figure 7, as the points marked by the same number can be in two districts which are far away from each other.

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.