Step by Step
T
Threshold — the firing point
A neuron "fires" (produces a significant output) once the total combined signal it receives exceeds a certain threshold, controlled by the activation function.
Example: like a real brain cell that only fires an electrical signal once enough incoming signals accumulate past a certain point.
W
Weights — the importance of each input
Each incoming signal is multiplied by a weight representing how important that particular input is to this neuron's decision — these weights are exactly what the network learns during training.
Example: one input signal might have a weight of 0.9 (very important to this neuron) while another has a weight of 0.05 (barely relevant).
A
Activation — deciding whether and how strongly to fire
After summing all weighted inputs, the activation function determines the neuron's actual output — whether it stays silent, passes the signal through, or fires with some intensity.
Example: like many weak nudges from different friends adding up to finally convince someone to go along with a plan, many small weighted signals can add up to cross the neuron's activation threshold.
Applied Walkthrough
1
A neuron receives five separate input signals, each individually too weak to matter much on its own.
2
Each signal gets multiplied by its own learned weight, reflecting how much that particular input matters to this neuron.
3
The weighted signals are summed together, and if that sum crosses the neuron's threshold (as determined by its activation function), the neuron fires.
4
This mirrors how a real brain cell works: weak individual signals from many different neurons can combine to trigger a response, even though no single input alone would have been enough.
Exam Application
Exams test whether you understand the biological inspiration behind artificial neurons (weak signals combining to cross a threshold) and whether you correctly identify that the weights — not the threshold or activation function itself — are what the network actually learns and adjusts during training.
⚠ Common Trap
The most common trap is thinking the threshold or activation function itself is what gets "trained." In most standard architectures, the activation function is a fixed, chosen hyperparameter — it's the weights (and biases) that are the learned parameters, adjusted through training to change how the neuron responds to its inputs.
✓ Quick Self-Check
1. What determines whether an artificial neuron "fires"?
Whether the sum of its weighted inputs exceeds a threshold, as determined by the activation function.
Tap to reveal / hide
2. What do the weights in a neuron represent?
How important each particular input is to that neuron's overall decision.
Tap to reveal / hide
3. What part of the neuron actually gets learned and adjusted during training?
The weights (and biases) — not typically the activation function itself, which is usually a fixed hyperparameter choice.
Tap to reveal / hide
4. What biological process is the artificial neuron's firing behavior modeled after?
A real brain cell (neuron), which fires an electrical signal once enough incoming signals from other neurons accumulate past a certain point.
Tap to reveal / hide
5. Can several individually weak input signals combine to make a neuron fire?
Yes — just as several weak signals from different "friends" (inputs) can combine to cross the firing threshold, even if no single input alone would be enough.
Tap to reveal / hide