Ablation (artificial intelligence)
In artificial intelligence (AI), particularly machine learning (ML), ablation is the removal of a component of an AI system. An ablation study studies the performance of an AI system by removing certain components, to understand the contribution of the component to the overall system. The term is by analogy with biology (removal of components of an organism), and, continuing the analogy, is particularly used in the analysis of artificial neural nets, by analogy with ablative brain surgery.[1] Ablation studies require that the system exhibit graceful degradation: that they continue to function even when certain components are missing or degraded.[2]
History
The term is credited to Allen Newell,[3] who used it in his 1974 tutorial on speech recognition, published in Newell (1975). The term is by analogy with ablation in biology. The motivation was that, while individual components are engineered, the contribution of an individual component to the overall system performance is not clear; removing components allows this analysis.[2]
References
- Meyes, Richard; Lu, Melanie; de Puiseau, Constantin Waubert; Meisen, Tobias (2019-01-24). "Ablation Studies in Artificial Neural Networks". arXiv:1901.08644. Cite journal requires
|journal=(help) - Newell 1975.
- Cohen & Howe 1988, p. 40, Ablation and substitution studies..
- Newell, Allen (1975). D. Raj Reddy (ed.). A Tutorial on Speech Understanding Systems. In Speech Recognition: Invited Papers Presented at the 1974 IEEE Symposium. New York: Academic. p. 43.
- Cohen, Paul R.; Howe, Adele E. (1988-12-15). "How Evaluation Guides AI Research: The Message Still Counts More than the Medium". AI Magazine. 9 (4): 35–43. doi:10.1609/aimag.v9i4.952. ISSN 2371-9621.