There are 4 metrics in classification problem, which is super vised learning.
- Accuracy
- prescision
- recall
- f1score
- confusion matrix.
Deep Learning:
Perceptron Model:
Inputs multiply by random weights, added a bias to avoid zero multiplication result, then activation function and then the output.
Activation Functions:
- Simple activation has output 0 for -ve and 1 for +ve.
- sigmoid output between 0 and 1
- tanh(x) output between -1 and 1
- Relu return max(0,x)
Cost Function: Quadratic Cost:
c = E(y-a)^2 / n it slow down in our learning speed
Cross Entropy:
This is for faster learning. The faster the learning rate as the difference between true value and prediction.
Gradient Descent and Back Propagation:
Gradient descent use to minimize the value of cost function. Back propagation calculate error contribution at each neuron after batch of data is processed.Its a methametical chain rule.It requires a known desired output of each input value.
YOLO V3 is trained model on COCO dataset of Microsoft, it has 80 different Classes YOLO
Audience:
- Beginners
- Intermediate
- Professionals