WEBVTT

00:00.000 --> 00:20.680
A forward pass is how a neural network makes a prediction. We feed numbers in. They flow

00:20.680 --> 00:27.720
forward through the layers. We read the answer at the end. Let's do one step by step.

00:27.720 --> 00:38.080
Our network has two inputs. X1 equals 2. X2 equals 1. These are the numbers we feed in.

00:38.080 --> 00:42.520
The weights and biases are already set. We're not training here. We're predicting. Our

00:42.520 --> 00:50.160
job is to follow the numbers forward. Let's compute neuron H1. X1 times its weight.

00:50.160 --> 01:01.040
2 times 0.5 equals 1.0. X2 times its weight. 1 times 1.0 equals 1.0. Add them together 2.0. Add

01:01.040 --> 01:13.720
the bias, which is 0, still 2.0. H1's pre-activation value is 2.0. Now we apply RLU. Is 2.0 positive?

01:13.720 --> 01:26.280
Yes. So we keep it. H1 outputs 2.0. Now neuron H2. X1 times 1.0. That's 2.0. X2 times negative 0.5. That's

01:26.280 --> 01:38.320
negative 0.5. Some 1.5. Add the bias of 1.0. Pre-activation is 2.5. RLU on 2.5. Positive. So we keep it. H2

01:38.320 --> 01:48.080
outputs 2.5. The hidden layer is done. H1 equals 2.0. H2 equals 2.5. Now the output neuron H1 value 2.0.

01:48.080 --> 02:02.080
Times its weight 2.0. That's 4.0. H2 value 2.5 times weight 1.0. That's 2.5. Some 6.5. Add bias 1.0. Pre-activation is 7.5.

02:02.080 --> 02:12.320
RLU of 7.5. Positive. We keep it. The network's prediction is 7.5. That is the output of 1

02:12.320 --> 02:21.440
full forward pass. Let's see the whole thing at once. Numbers in at the left. Multiply by weights. Add

02:21.440 --> 02:28.640
the bias. Apply RLU. Pass forward. Do it again at the next layer. Read the answer at the end. That is a

02:28.640 --> 02:30.640
forward pass.