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 using the symmetrical MSAF. The results contain positive states and zero states as well as negative states.
© Copyright Policy
Related In: Results  -  Collection

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

fig9: Results of classifying random points using the symmetrical MSAF. The results contain positive states and zero states as well as negative states.

Mentions: Using the symmetrical MSAF, the neural network is able to classify negative values and output negative states as well. For instance, let the activation function be y = −1 + 1/(1 + e−x−20) + 1/(1 + e−x) and (w1, w2, w3, w4, w5, w6, b1, b2, b3) = (−24, −24,24,24, −40,32,16, −32,24); then the network will classify random points shown in Figure 9. It is easy to note that the inputs and the hidden units both contain negative coordinates. In fact, the symmetrical MSAF enables the neural network to have positive and negative states at the same time.


Deep Neural Networks with Multistate Activation Functions.

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

Results of classifying random points using the symmetrical MSAF. The results contain positive states and zero states as well as negative states.
© Copyright Policy
Related In: Results  -  Collection

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

fig9: Results of classifying random points using the symmetrical MSAF. The results contain positive states and zero states as well as negative states.
Mentions: Using the symmetrical MSAF, the neural network is able to classify negative values and output negative states as well. For instance, let the activation function be y = −1 + 1/(1 + e−x−20) + 1/(1 + e−x) and (w1, w2, w3, w4, w5, w6, b1, b2, b3) = (−24, −24,24,24, −40,32,16, −32,24); then the network will classify random points shown in Figure 9. It is easy to note that the inputs and the hidden units both contain negative coordinates. In fact, the symmetrical MSAF enables the neural network to have positive and negative states at the same time.

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.