AI System Ablation Study

From GM-RKB
(Redirected from Ablation study)
Jump to navigation Jump to search

An AI System Ablation Study is an AI system analysis that systematically disables AI system components to understand how each component contributes to the overall system's performance.



References

2024

  • (Wikipedia, 2024) ⇒ https://en.wikipedia.org/wiki/Ablation_(artificial_intelligence) Retrieved:2024-8-8.
    • In artificial intelligence (AI), particularly machine learning (ML), ablation is the removal of a component of an AI system. An ablation study investigates the performance of an AI system by removing certain components to understand the contribution of the component to the overall system. The term is an analogy with biology (removal of components of an organism), and is particularly used in the analysis of artificial neural nets by analogy with ablative brain surgery. Other analogies include other neuroscience biological systems such as Drosophila central nervous system and the vertebrate brain. Ablation studies require that a system exhibit graceful degradation: the system must continue to function even when certain components are missing or degraded. According to some researchers, ablation studies have been deemed a convenient technique in investigating artificial intelligence and its durability to structural damages. Ablation studies damage and/or remove certain components in a controlled setting to investigate all possible outcomes of system failure; this characterizes how each action impacts the system's overall performance and capabilities. The ablation process can be used to test systems that perform tasks such as speech recognition, visual object recognition, and robot control.

2024

  • (Wikipedia, 2024) ⇒ https://en.wikipedia.org/wiki/Ablation_(artificial_intelligence)#History Retrieved:2024-8-8.
    • The term is credited to Allen Newell,one of the founders of artificial intelligence, who used it in his 1974 tutorial on speech recognition, published in . 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. Newell compared the human brain to artificial computers. With this in thought, Newell saw both as knowledge systems whereas procedures such as ablation can be performed on both to test certain hypotheses.

2022

  • (Hameed et al., 2022) ⇒ Isha Hameed, Samuel Sharpe, Daniel Barcklow, Justin Au-Yeung, Sahil Verma, Jocelyn Huang, Brian Barr, and C. Bayan Bruss. (2022). "Based-xai: Breaking ablation studies down for explainable artificial intelligence." arXiv preprint arXiv:2207.05566.
    • ABSTRACT: "Explainable artificial intelligence (XAI) methods lack ground truth. In its place, method developers have relied on axioms to determine desirable properties for their explanations' behavior. For high stakes uses of machine learning that require explainability, it is not sufficient to rely on axioms as the implementation, or its usage, can fail to live up to the ideal. As a result, there exists active research on validating the performance of XAI methods. The need for validation is especially magnified in domains with a reliance on XAI. A procedure frequently used to assess their utility, and to some extent their fidelity, is an ablation study. By perturbing the input variables in rank order of importance, the goal is to assess the sensitivity of the model's performance. Perturbing important variables should correlate with larger decreases in measures of model capability than perturbing less important features. While the intent is clear, the actual implementation details have not been studied rigorously for tabular data. Using five datasets, three XAI methods, four baselines, and three perturbations, we aim to show 1) how varying perturbations and adding simple guardrails can help to avoid potentially flawed conclusions, 2) how treatment of categorical variables is an important consideration in both post-hoc explainability and ablation studies, and 3) how

2021

  • (Sheikholeslami et al., 2021) ⇒ Sina Sheikholeslami, Moritz Meister, Tianze Wang, Amir H. Payberah, Vladimir Vlassov, and Jim Dowling. (2021). "Autoablation: Automated parallel ablation studies for deep learning." In: Proceedings of the 1st Workshop on Machine Learning and Systems, pp. 55-61.
    • ABSTRACT: Ablation studies provide insights into the relative contribution of different architectural and regularization components to machine learning models' performance. In this paper, we introduce AutoAblation, a new framework for the design and parallel execution of ablation experiments. AutoAblation provides a declarative approach to defining ablation experiments on model architectures and training datasets, and enables the parallel execution of ablation trials. This reduces the execution time and allows more comprehensive experiments by exploiting larger amounts of computational resources. We show that AutoAblation can provide near-linear scalability by performing an ablation study on the modules of the Inception-v3 network trained on the TenGeoPSAR dataset.