Rumelhart, Hinton & Williams, Learning representations by back-propagating errors, Nature 1986 · p. 535, fig. 4 caption
Saturation & the Tolerance Window
Why the paper counted any output above 0.8 or below 0.2 as correct
The logistic and its slope
drag the probe along x
probe target d = 0
d = 1
One unit, two inputs
x = w₁i₁ + w₂i₂ + θ · learning: Δw = −ε ∂E/∂w
Start
Step once
Reset
cases
i₁
i₂
d = 0
d = 1
start
w₁
w₂
θ
learning
ε
linear time
log time