Classical machine learning algorithms often assume that the data are drawn\ni.i.d. from a stationary probability distribution. Recently, continual learning\nemerged as a rapidly growing area of machine learning where this assumption is\nrelaxed, i.e. where the data distribution is non-stationary and changes over\ntime. This paper represents the state of data distribution by a context\nvariable $c$. A drift in $c$ leads to a data distribution drift.\n A context drift may change the target distribution, the input distribution,\nor both. Moreover, distribution drifts might be abrupt or gradual. In continual\nlearning, context drifts may interfere with the learning process and erase\npreviously learned knowledge; thus, continual learning algorithms must include\nspecialized mechanisms to deal with such drifts. In this paper, we aim to\nidentify and categorize different types of context drifts and potential\nassumptions about them, to better characterize various continual-learning\nscenarios. Moreover, we propose to use the distribution drift framework to\nprovide more precise definitions of several terms commonly used in the\ncontinual learning field.\n