MLP, CNN and RNN

Model	Connection Style	Best For	Weakness
MLP	Fully connected, dense layers	Tabular data, general-purpose learning	Ignores structure (order, spatial locality)
CNN	Convolutions (local filters)	Images, spatial data	Struggles with sequences without modification
RNN	Sequential recurrence	Time series, language	Hard with long dependencies, slow training