Weights control how strongly each input influences a neuron.
The bias shifts the neuron's output up or down independently of the input.
Together they are the parameters the network learns.
They start as small random numbers before training begins.
Training is nothing more than repeatedly adjusting these two things.
Weights and biases are the numbers the network learns, and there is nothing else in there.
A weight sits on a connection and says how much that input matters to this neuron.
A large positive weight means the input pushes the neuron hard towards firing.
A negative weight means it pushes the other way, which is just as useful.
A weight near zero means the network decided that input barely matters here.
The bias is a single number added after the weighted sum.
It shifts the threshold, so the neuron can fire readily or only under protest.
Without a bias every neuron is forced to make its decision through the origin.
A layer with 100 inputs and 50 neurons holds 5,000 weights and 50 biases.
Stack a few of those and the count reaches millions without trying.
They start as small random numbers, because starting them equal leaves them equal.
Training adjusts every one of them a fraction at a time.
Nothing about a single trained weight is meaningful on its own.
The knowledge is spread across all of them together, which is why models are hard to read.
When someone says a model has seven billion parameters, this is what they are counting.
One question on this card. Every option explains itself once you pick, right or wrong.
25 concepts in this module, then a graded test with every answer explained.