Neural networks with periodic and monotonic activation functions: a comparative study in classification problems

Predicting access times is a crucial part of predicting hard disk drive performance. Existing approaches use white-box modeling and require intimate knowledge of the internal layout of the drive, which can take months to extract. Automatically learning this behavior is a much more desirable approach, requiring less expert knowledge, fewer assumptions, and less time. Others have created behavioral models of hard disk drive performance, but none have shown low per-request errors. A barrier to machine learning of access times has been the existence of periodic behavior with high, unknown frequencies. We show how hard disk drive access times can be predicted to within 0.83 ms using a neural net after these frequencies are found using Fourier analysis.

show abstract

“…Lack of a such a limit can lead to instability [31]. Periodic activation functions also introduce many local minima [28].…”

Section: Learning Periodic Functionsmentioning

confidence: 99%

Fourier-assisted machine learning of hard disk drive access time models

Crume

Maltzahn

Ward

et al. 2013

Proceedings of the 8th Parallel Data Storage Workshop

View full text Add to dashboard Cite

show abstract

“…We also constructed an MLP with a hidden layer of 24 sinusoidal units, as in [34]. Initial frequencies for BPW were randomly assigned to an interval [ − 3:5; 3:5], and the initial range for the coe cients was 0:0001.…”

Section: The Two Spirals Problemmentioning

confidence: 99%

Margin maximization with feed-forward neural networks: a comparative study with SVM and AdaBoost

2004

View full text Add to dashboard Cite

“…Multiple valued neuron appears in the activation function and neural network where input and output value are both complex number with unitary length and this directly leads to consider sine and cosine function as the activation function in this case. And people have tried to compare training time between the periodic function and other activation function such as sigmoid [11]. It's interesting to see how the periodic functions are compared to more recently used activation function such as linear rectifier function.…”

Section: Introductionmentioning

confidence: 99%

Periodic Function as Activation Function for Neural Networks

Xu¹,

Guan²,

Cai³

2017

dtcse

View full text Add to dashboard Cite

Abstract. In this paper, we explore the periodic function as alternative activation function for neural network. Previously sigmoid function is used as standard activation function for neuron and now linear rectifier function are used. Even max-out function can be learned as a general form convex activation function. We explore the possibility in the other direction, where we use periodic function as activation function. We expect network with less layer and less neuron can capture the target distribution. The experiments verify our expectation and show that period function can act as an alternative activation function.

show abstract

Neural networks with periodic and monotonic activation functions: a comparative study in classification problems

Cited by 41 publications

References 0 publications

Fourier-assisted machine learning of hard disk drive access time models

Fourier-assisted machine learning of hard disk drive access time models

Margin maximization with feed-forward neural networks: a comparative study with SVM and AdaBoost

Periodic Function as Activation Function for Neural Networks

Contact Info

Product

Resources

About