🧠 Neural Network Calculator
Enter the neurons in each layer, input to output, and see exactly how many weights, biases, and total parameters your fully-connected network holds.
🧮 Count Your Network's Parameters
What is a Neural Network Calculator?
Every neuron in a dense network connects to every neuron in the next layer, and each connection carries a learnable weight. This calculator adds up all those weights, plus one bias per neuron, to reveal the total number of parameters — the numbers a network adjusts as it learns.
It turns an intimidating idea into simple arithmetic you can watch. Add a layer, make one wider, and see the parameter count jump — a vivid lesson in why model size, memory, and data requirements all grow together.
❓ Frequently Asked Questions
How do you count parameters in a neural network?
For a fully-connected (dense) network you walk through each pair of adjacent layers. Between a layer of a neurons and the next layer of b neurons there are a × b weights — one for every connection — plus b biases, one per neuron in the second layer. Add those up across every pair of layers and you get the total number of trainable parameters. This calculator does that arithmetic for the layer sizes you enter.
What's the difference between weights and biases?
A weight scales the signal travelling along one connection between two neurons, so the number of weights equals the number of connections. A bias is a single extra value added to each neuron before its activation function, letting the neuron shift its output up or down independently of its inputs. Weights vastly outnumber biases, which is why the weight count dominates the total parameters.
Why does the parameter count matter?
Parameters are what the network learns, so their number is a rough measure of a model's capacity. More parameters can capture more complex patterns, but they also need more memory to store, more compute to train, and more data to fit without overfitting. Seeing how quickly the count grows when you widen or deepen a network builds intuition for why bigger isn't automatically better.
Does this cover convolutional or transformer networks?
This calculator models fully-connected (dense) layers, the classic building block taught first and used in the final layers of many networks. Convolutional and transformer layers share parameters in structured ways and count differently — a small convolutional kernel can process a huge image with very few weights. Use this tool to master the dense case, which is the foundation for understanding those more advanced architectures.