A Parameter in machine learning and AI is a learnable variable internal to a model whose values are adjusted during training to minimise a loss function, as distinguished from hyperparameters, which are configuration choices set before training begins. In neural networks, parameters encompass weight matrices and bias vectors in each layer; the total parameter count (ranging from thousands in small models to hundreds of billions in large language models) is a primary indicator of model capacity and computational cost. Parameters are initialised randomly or via transfer learning, then updated iteratively through gradient-based optimisation algorithms such as stochastic gradient descent, encoding learned representations of the training distribution.
Semantic Classification
Content
Parameter — content pending enrichment.