Tensor

A tensor is a mathematical container for numbers arranged across one or more dimensions. A scalar can be treated as a zero-dimensional tensor, a list as one-dimensional, a table as two-dimensional, and images or batches of data as higher-dimensional tensors. Deep learning frameworks use tensors to store model inputs, parameters, activations, and gradients while performing efficient operations on CPUs, GPUs, or specialized accelerators. A tensor’s shape, data type, device, and memory layout determine which operations are valid and how much computation is required. In some mathematical contexts the word has a stricter meaning, but in everyday machine learning it usually refers to a multidimensional numerical array.