What Is Beam Search in AI?
Beam search keeps several promising partial sequences at each generation step instead of committing to only one next token.
Don't Just Read it – Explore It
Beam search keeps several promising partial sequences at each generation step instead of committing to only one next token.
An autoencoder learns to compress an input into a representation and reconstruct the original input from that representation.
A vision transformer represents an image as a sequence of patches and processes those patch representations with transformer layers.
A long short-term memory network is a recurrent architecture that uses gates and a cell state to preserve or replace sequence information.
An encoder-decoder architecture converts an input into an internal representation and uses that representation to generate an output.
A causal language model predicts each token using only earlier tokens in the sequence, supporting left-to-right generation.
Positional encoding supplies order information to models whose attention operations do not inherently know token positions.
Gradient clipping limits the size of gradients or parameter updates to reduce instability caused by exploding gradients.
A residual connection adds an earlier representation to a later layer, creating a shortcut through a neural network.
Batch normalization standardizes intermediate activations using mini-batch statistics and then learns a scale and offset.
A learning rate controls how large each parameter update is during gradient-based model training.
Dropout is a regularization method that randomly suppresses some activations during training to reduce reliance on particular units.