Architecture Playground

Visually understand how neural network architectures work, how parameters scale, memory changes, and compute grows. See the real math behind the models.

MLP Configuration

Input Size (Features)3

Hidden Layers2

Hidden Size (Width)4

Output Size (Classes)2

Weight Precision

Total Parameters

36 Weights + 10 Biases

Compute (FLOPs)

per Forward Pass

Train Step: 216

Storage Memory

184 B

At FP32 precision

MLP Deep Dive

All the math behind your 3→4×2→2 network — live

A Multi-Layer Perceptron is a stack of fully-connected layers. Every neuron in one layer sends a signal to every neuron in the next — that is why it's called fully connected.

Input (3)→Hidden 1 (4)→Hidden 2 (4)→Output (2)

Layer	Neurons	Weights	Biases	Subtotal	Math
Input Layer	3	0	0	0	Raw features — no params
Hidden 1	4	12	4	16	z = Σ(w·x) + b → a = σ(z)
Hidden 2	4	16	4	20	z = Σ(w·x) + b → a = σ(z)
Output Layer	2	8	2	10	Softmax → class probabilities
Grand Total				46	36 W + 10 B