INTERNATIONAL JOURNAL OF SCIENTIFIC DEVELOPMENT AND RESEARCH International Peer Reviewed & Refereed Journals, Open Access Journal ISSN Approved Journal No: 2455-2631 | Impact factor: 8.15 | ESTD Year: 2016
open access , Peer-reviewed, and Refereed Journals, Impact factor 8.15
When training and building a neural network, a number of subtle but important decisions needs to be taken. Zeroing down on the loss function to be used, the number of layers, kernel size, and the stride for each convolution layer, best-suited optimization algorithm for the network, etc. Compared to all these things, the choice of initialization of weights may seem trivial pre-training detail. But weight initialization contributes as a significant factor on the final quality of a network as well as its convergence rate. This paper discusses different approaches to weight initialization and compares their results on few datasets to find out the best technique that can be employed to achieve higher accuracy in relatively lower duration.
Keywords:
Neural Networks, Variable Initialization
Cite Article:
"Initialization of Weights in Neural Networks", International Journal of Science & Engineering Development Research (www.ijsdr.org), ISSN:2455-2631, Vol.3, Issue 11, page no.73 - 79, November-2018, Available :http://www.ijsdr.org/papers/IJSDR1811013.pdf
Downloads:
000337070
Publication Details:
Published Paper ID: IJSDR1811013
Registration ID:180734
Published In: Volume 3 Issue 11, November-2018
DOI (Digital Object Identifier):
Page No: 73 - 79
Publisher: IJSDR | www.ijsdr.org
ISSN Number: 2455-2631
Facebook Twitter Instagram LinkedIn