Novelty detection is the
mechanism by which an
intelligentorganism is able to identify an incoming
sensory pattern as being hitherto unknown. If the pattern is sufficiently
salient or associated with a high positive or strong negative
utility, it will be given computational resources for effective future processing.
The principle is long known in
neurophysiology, with roots in the
orienting response research by
E. N. Sokolov[1] in the 1950s. The reverse phenomenon is
habituation, i.e., the phenomenon that known patterns yield a less marked response. Early neural modeling attempts were by Yehuda Salu.[2] An increasing body of knowledge has been collected concerning the corresponding mechanisms in the brain.[3][4] In technology, the principle became important for
radar detection methods during the Cold War, where unusual aircraft-reflection patterns could indicate an attack by a new type of aircraft. Today, the phenomenon plays an important role in
machine learning and
data science, where the corresponding methods are known as
anomaly detection or outlier detection. An extensive methodological overview is given by Markou and Singh.[5][6]
^Sokolov, E.N. (1960). "Neuronal models and the orienting reflex". The Central Nervous System and Behavior. Josiah Macy, Jr. Foundation. pp. 187–276.
OCLC222201512.
^Markou, M.; Singh, S. (2003). "Novelty detection: a review — Part 1: statistical approaches". Signal Processing. 83 (12): 2481–97.
doi:
10.1016/j.sigpro.2003.07.018.
S2CID17490415.
^Markou, M.; Singh, S. (2003). "Novelty detection: a review — Part 2: neural network based approaches". Signal Processing. 83 (12): 2499–2521.
doi:
10.1016/j.sigpro.2003.07.019.