Neural network Algos formulas
Download as DOC, PDF0 likes307 views
The document discusses various neural network learning algorithms, including Hebbian learning, Perceptron, Adaline, and Hopfield networks. It outlines the processes for weight updates, activation functions, and stopping conditions for both feed-forward and unsupervised learning scenarios. The focus is on defining how weights are adjusted based on sample input, error calculation, and convergence criteria.
![Self unsupervised Dj=∑(wij-xi)2 Choose the Wij(new)= Wij(old)+α[xi-wij(old)]
Organization learning minimum Dj (new)= 0.5 α (old)
map and set the If
Feed- value of j convergence
Forward according to criterion met,
it. STOP.
Or
When cluster
1 and cluster
2 is inverse of
each other.](https://image.slidesharecdn.com/neuralnetworkformulae-130222123136-phpapp01/85/Neural-network-Algos-formulas-2-320.jpg)
Neural network Algos formulas
- 1. Algorithm Architectur Net Input Activation Weight Update Stopping name e Function Condition AND Biased update Hebb-Net Single Layer, wij(new)=wij(old)+xi y Only one Feed-forward - - bj(new)=bj(old)+y iteration Perceptrone Dual layer y_in= bj+Σxiwij y= 1 if y_in>θ wij(new)=wij(old)+ α txi If y=t, for all Feed-forward y= 0 if bj(new)=bj(old)+ α t samples -θ ≤y_in≤ θ where t=target y= -1 if y_in< -θ Adaline Feed-forward y_in=Σxiwi + b y= 1 if y_in≥ θ b(new)=b(old)+α(t-y_in), If the greatest y=-1 if y_in < θ wi(new)=wi(old)+α(t-y_in)xi weight change is smaller then the applied threshold. Madaline Dual Layer Z_inj=bj+∑xi wij f (x)=1 if x>0 When t=-1 If weight y_in=b3+z1v1+z2v2 -1 if x<0 bj(new)=bj(old)+α(-1-z_inj), changes have wij(new)=wij(old)+ α(-1- stopped so z_inj)xi one Iteration is complete when t=1 bj(new)=bj(old)+α(1-z_inj), wij(new)=wij(old)+ α(1- z_inj)xi Hetero Single Layer Y_inj=Σxiwij Yj=1 if y_inj>θj All samples Associative Yj if y_inj= θj wij(new)=wij(old)+sitj have been -1 if y_inj< θj processed Auto Single Layer Y_inj=Σxiwij Yj=1 if y_inj>0 All samples Associative -1 if y_inj<0 wij(new)=wij(old)+xiyj have been processed Discrete Unsupervised Y_inj = xi + Σyiwji 1 if y-ini> θ Hopfield Learning yi if y-ini= θ 0 if y-ini< θ Feedbackward Back Multi-layer Wij(new) = wij(old) + α Errj Oi We will solve propagation supervised Y_inj = Σwijxi + bj bj = bj(old) + α Errj it until the learning Yj=1/1-e-Y_in error is zero Errors: Err=0 Feed-forward For hidden layers Errj = Oj (1-Oj) ∑ Errk wjk For output layer Errj = Oj (1-Oj) (Tj-Oj)
- 2. Self unsupervised Dj=∑(wij-xi)2 Choose the Wij(new)= Wij(old)+α[xi-wij(old)] Organization learning minimum Dj (new)= 0.5 α (old) map and set the If Feed- value of j convergence Forward according to criterion met, it. STOP. Or When cluster 1 and cluster 2 is inverse of each other.