Forward Propagation in Neural Networks | Deep Learning
Satyajit Pattnaik
https://www.youtube.com/watch?v=99CcviQchd8

Transkript (automatisch erstellt)

0:00 Let's learn about forward propagation in neural network. We have three layers in a neural network, input, hidden, and output.
0:11 Hidden layer is the layer where the learning happens. Each neuron is activated based on activation functions.
0:20 Forward propagation is the movement of information from input to the output layer via the hidden layers. At a very high level, this is how the
0:31 information is traversed. Each neuron in first hidden layer does a dot product of the weight and input they receive.
0:42 Then this total is fed to an activation function. And the output of which goes to the next hidden layers.
0:51 This flow repeats until it hits the output layer. In output neuron, activation function is different from the activation functions
1:02 used in the hidden layer neurons. This activation functions depends on the task which we are doing. For example, if we are doing a binary
1:13 classification, we can use a sigmoid function at the output layer. If activation is greater than 0.5, then
1:23 one, else zero because sigmoid function ranges between zero and one. This flow of information change depends on the architecture we use.
1:35 But the general idea is same. Let's see how forward propagation works for a single data point. This is a program to understand the
1:46 concept. We have two nodes of a hidden layer. We have an input which is one and two. We have defined the weights at each
1:57 node. Value at node zero is basically input data into node zero weights. Similarly, value at node one is basically input
2:09 data multiplied with node one weights. Similarly, hidden layer results is done by adding value one and value two. And then it is multiplied with output
2:22 weights to find the final output. That's all about forward propagation. In the next video, we will learn about back
2:33 propagation in neural networks.