A Conditional Random Field (CRF) is a probabilistic model often used for structured prediction in sequence labeling tasks (like part-of-speech tagging or image segmentation boundaries). It models the conditional probability of an entire output sequence (or label field) given an input sequence, capturing dependencies between output labels (unlike independent classifiers). Technically, a CRF is an undirected graphical model (a kind of Markov Random Field) where each node is conditioned on observed data. CRFs compute the most likely label configuration via algorithms like the Viterbi algorithm or belief propagation. They excel in tasks where context matters, ensuring that label predictions are globally consistent (e.g., in an image, neighboring pixels’ labels influence each other to produce coherent segmentations)..