Weight - Page 2

A weight is a numerical parameter that controls how strongly one signal influences another inside a machine learning model. During neural network training, optimization algorithms adjust millions or billions of weights so the model’s output better matches the desired result. Positive, negative, and near-zero values can strengthen, reverse, or reduce a connection’s effect. Individual weights usually have little meaning on their own because behavior emerges from many interacting parameters across layers. Model size is often described by its parameter count, much of which consists of weights. Training, fine-tuning, pruning, quantization, and regularization all change or represent weights in different ways.