Translation invariance is a property of a system or function where its behavior or output does not change when inputs are shifted uniformly in space. This concept is fundamental in fields like physics and machine learning, particularly in convolutional neural networks, where it allows for the recognition of patterns regardless of their position in the input data.