How to read these. Each diagram is a simplified sketch. The purpose is not architectural precision but intuition: how information flows and what the network is built for. Examples show the kind of task each network is commonly used to solve.
Perceptron (single layer)
Inputs connect directly to output. Example: a simple yes/no classification task.
Feedforward (multi-layer)
Hidden layer allows non-linear patterns. Example: XOR or any task where a straight line will not separate categories.
Convolutional Network (CNN)
Small filters scan across space. Example: detecting edges and then faces in images.
Recurrent Network (RNN)
State is fed forward in time. Example: predicting the next word in a sentence.
Transformer (attention)
Each token can attend to distant tokens. Example: summarizing a paragraph by using context from many words at once.
Reinforcement Learning
Agent acts, environment returns reward. Example: learning to navigate a maze or play a game.