The forward pass: how a network turns inputs into a prediction

Before a network can learn anything, it has to make a guess. Let's trace the exact path from numbers in to one number out.

3 quick questions · about 2 min · no sign-up

Question 1 of 3

A network takes two inputs and each one arrives with its own weight. What's the very first thing it does with an input and its weight?

• Multiplies the input by its weight
• Adds the weight to the input
• Compares the input to the weight to see which is bigger
• I'm not sure

You said: Multiplies the input by its weight

Exactly. Each input gets scaled by its own weight: input times weight. The weight is how much that input is allowed to matter. That single multiplication is the basic move repeated across every connection.

You said: Adds the weight to the input

It's multiply, not add. Each input is scaled by its weight: input times weight. The weight controls how much that input counts, and scaling is what makes a big weight amplify an input and a tiny weight mute it.

You said: Compares the input to the weight to see which is bigger

No comparison happens here. The network multiplies the input by its weight: input times weight. The weight scales the input up or down, deciding how much that input contributes.

You said: I'm not sure

No problem. The first move is multiplication: each input is multiplied by its own weight. The weight decides how much that input matters, so input times weight scales it up or down.

Another way to see it

Think of each weight as a volume knob on one input. Turning the knob up multiplies that input by a bigger number so it dominates; turning it down toward zero multiplies it small so it barely registers. The network's first act is just turning all the knobs.

So every input becomes input times weight. Now what happens to all those scaled numbers?

Question 2 of 3

After each input is multiplied by its weight, what does the network do with those scaled values to move toward a single prediction?

• Adds them all together into one sum
• Keeps each scaled value separate as its own output
• Multiplies the scaled values together
• I'm not sure

You said: Adds them all together into one sum

Right. The scaled inputs are summed into a single number. Multiply each input by its weight, then add the results: that running total is what gets passed on to become the prediction.

You said: Keeps each scaled value separate as its own output

They don't stay separate. The scaled inputs get added together into one sum. That collapse from many numbers down to one total is exactly how the network heads toward a single prediction.

You said: Multiplies the scaled values together

Close on the idea of combining, but it's addition, not multiplication. The weighted inputs are summed: multiply each input by its weight, then add those products into one total that becomes the prediction.

You said: I'm not sure

All good. After scaling, the network adds the weighted inputs together into one sum. Multiply each input by its weight, then total them up: that single sum is what gets passed forward.

Multiply, then sum. Let's see if you can run the whole path on real numbers.

Question 3 of 3

Inputs are 2 and 3. Their weights are 5 and 10. Run the forward pass: scale each input, then sum. What number does the network pass forward as its prediction?

• 40
• 20 (added the inputs and weights: 2+3+5+10)
• 300 (multiplied the two products: 10 x 30)
• I'm not sure

You said: 40

Exactly right: 2 times 5 is 10, 3 times 10 is 30, and 10 plus 30 is 40. You just ran a full forward pass: scale each input by its weight, then sum. That 40 is the network's raw guess.

You said: 20 (added the inputs and weights: 2+3+5+10)

That summed everything without weighting. You multiply first: 2 times 5 is 10, 3 times 10 is 30, then sum to get 40. Scale each input by its own weight, then add: the answer is 40.

You said: 300 (multiplied the two products: 10 x 30)

You scaled correctly (2 times 5 is 10, 3 times 10 is 30) but then multiplied instead of adding. The final step is a sum: 10 plus 30 is 40. So the network passes forward 40.

You said: I'm not sure

Let's walk it: 2 times 5 is 10, 3 times 10 is 30, then add for 10 plus 30 equals 40. Scale each input by its weight, then sum: the prediction is 40.

The takeaway

The forward pass is just two moves: multiply each input by its weight, then add the results into one number. That single number is the network's guess, made with no learning yet.

Next step

You can now turn inputs into a prediction — but how do we know if that prediction is any good? Next: scoring the error with a loss.