Neural networks are a class of models inspired by the human brain’s interconnected neurons, comprised of layers of simple processing units (neurons). Each neuron computes a weighted sum of its inputs and applies a non-linear activation function. Networks have an input layer (features), one or more hidden layers (that learn intermediate representations), and an output layer (predictions). Feedforward neural networks (multi-layer perceptrons) connect each layer to the next without loops; recurrent neural networks (RNNs) have connections that form cycles for sequence processing; convolutional neural networks (CNNs) leverage local connectivity for grid data like images; transformers use attention mechanisms for sequence data. Neural nets are trained via backpropagation, adjusting weights to minimize a loss function. They’re known for being universal function approximators — given enough neurons, they can model very complex functions. Deep neural networks (with many layers) have achieved state-of-the-art results in many fields (vision, speech, NLP, etc.), at the cost of requiring a lot of data and compute, and being less interpretable.