ºÝºÝߣ

ºÝºÝߣShare a Scribd company logo

Convolutional Neural Network¤òÓ䤤¿ÖªÒ™½»Ìæ¤ÎÕJ×R¥â¥Æ?¥ë

?
?
K
Koki Nakamura

This document discusses using a convolutional neural network (CNN) for perceptual change recognition. It describes using a pre-trained VGG16 CNN model and fine-tuning the fully connected layers. The CNN is able to learn border-ownership representations from unlabeled data and can recognize changes between ambiguous figures with over 80% accuracy. Training takes around 158 simulation steps, with recognition performance improving over time.

1 of 16
Download to read offline
Convolutional Neural Network Recognition Model of Perceptual Change Using Convolutional Neural Network 2016 12 15
? ? ? ? [1] ? ? ? 2 [1]Rock, I., Hall, S., Davis, J..: Why do ambiguous ?gures reverse?, Acta Psychologica, Vol. 87, pp. 33¨C 59 (1994)
? [2] ? ? [3] ? border-ownership 3 ? [2] , : , D, Vol. 84, No. 7, pp. 1485¨C1494 (2001) [3] , : , , Vol. 103, No. 732, pp. 117¨C122 (2004)
? ? ? ? ? 4 Convolutional Neural Network
5 ? Convolutional Neural Network(CNN) ? [4] ? CNN VGG16 ? full connect ?ne tuning ? 512¡Á7¡Á7 [4]S. Lawrence and C. L. Giles and Ah Chung Tsoi and A. D. Back: Face Recognition: A Convolutional Neural-Network Approach, IEEE Transactions on Neural Networks, Vol. 8, No. 1, pp. 98¨C113 (1997) VGG16 (convolution , pooling ) VGG16 (full connect ) 0.8 0.2
6 ? / ? ? ? ? 512¡Á1¡Á1
? ? ? 7 0.8 0.2 0.20.8
? ? ? ? ? 8 512¡Á7¡Á7 0.9 0.1 512¡Á1¡Á1 ¡Á
? ? ? ? 9 512¡Á7¡Á7 512¡Á1¡Á1 ¡Á
? ? ? ? ? ? ? ? , ? ? 158 ? 10
11 time (simulation step) time (simulation step) time (simulation step)
? ? ? / ? ? / ? 2 ? / ? ? zeiler [4] ? 12 [5]Zeiler, Matthew D., Fergus, R.: Visualizing and un- derstanding convolutional networks, European Con- ference on Computer Vision, pp. 818¨C833 (2014)
13 A B ?
14 C ? ? C D ?
15 Convolutional Neural Network ? ? ? ? ?
Convolutional Neural Network¤òÓ䤤¿ÖªÒ™½»Ìæ¤ÎÕJ×R¥â¥Æ?¥ë

More Related Content

Convolutional Neural Network¤òÓ䤤¿ÖªÒ™½»Ìæ¤ÎÕJ×R¥â¥Æ?¥ë