X hits on this document

PDF document

The split-fovea model of visual word - page 55 / 78

213 views

0 shares

0 downloads

0 comments

55 / 78

Chapter 2. The Simulations

50

Figure 2.4: Control net: MSE at every fixation position

Figure 2.5: Fixation net: MSE at every fixation position

noted, fixations slightly left of centre seem to be easier for the control to learn. This can again be seen at epoch 300 where positions 3 and 2 have the lowest MSEs (in that order). A similar picture can again be seen at the end of training in figures 2.8 and 2.9. The fixation net has the lowest MSEs at positions 4 and 3 which corresponds to its fixation position frequencies and the control net has the lowest MSE’s for positions 3 and 2 confirming the finding of the previous model with the complete training set. Furthermore, the curves are further apart for the fixation net, with position 1 being much worse than the rest because it has the lowest frequency.

Document info
Document views213
Page views222
Page last viewedWed Dec 07 23:51:46 UTC 2016
Pages78
Paragraphs3087
Words17414

Comments