Construct · 6 fields · 14 citations
One name. Different field. Different construct. One map.
● jingle · one name, six constructs Nearest construct by meaningbreakdowncompare the meaningsEvery construct on the map is compared with every other by the meaning of its definitions, and the closest one is surfaced here. The nearest construct to robustness is breakdown.
Hold this as a prompt, not a verdict. Two constructs sitting close in meaning is a reason to read them side by side and decide for yourself whether they are one idea under two names. It is not proof that they are. Compare the meanings in the Collider.
Spokes are fields; dashed arcs are shared lineage. Tap a field for its definition, unit, and measures.
Drag to watch the word get claimed. Each field appears at the publication date of the earliest source cited for it. Tap a year to see which.
A model's ability to keep its accuracy when inputs are perturbed or drawn from a distribution other than the training one. Szegedy and colleagues found in 2014 that imperceptible pixel changes flip a network's label; Goodfellow, Shlens, and Szegedy explained the effect by the model's near-linearity and gave the one-step fast gradient sign attack; Madry and colleagues in 2018 recast defence as a min–max problem, train on the worst-case perturbation within a small ball. Hendrycks and Dietterich's ImageNet-C benchmark measures the natural version, accuracy under blur, noise, and weather, and Taori and colleagues showed in 2020 that most synthetic robustness fails to predict robustness to real distribution shift. Two constructs share the name: a worst-case guarantee inside an ℓp ball, and average accuracy on a different dataset.[1][2][3]
An estimator's ability to give nearly the right answer when the data depart from the assumed distribution, and in particular when a few observations are contaminated. Tukey's 1960 sampling study showed that a small fraction of wide-tailed contamination destroys the efficiency of the sample mean. Huber's 1964 paper founded the theory: minimise the maximum asymptotic variance over a neighbourhood of the normal, which yields the estimator that behaves like the mean in the middle and the median in the tails. Hampel's 1974 influence function measures the effect of one observation on the estimate; its supremum is the gross-error sensitivity, and the breakdown point is the contamination fraction that sends the estimate to infinity. Robustness here is a property of a procedure against a specified neighbourhood of models, not against an attacker.[4][5][6]
A closed-loop system's guarantee of stability and performance for every plant in a bounded set around the nominal model, since the model is never exact. Doyle's 1978 two-page paper showed that optimal LQG regulators can have arbitrarily small gain and phase margins (an optimal controller with no robustness) and the field reorganised around that result. Zhou, Doyle, and Glover's 1996 textbook fixed the machinery: unstructured and structured uncertainty, small-gain conditions, H∞ synthesis, and the structured singular value μ for the worst case over the uncertainty set. Robustness is a certificate computed before the system runs, and the disturbance is a set of possible plants rather than a set of possible data.[7][8]
The maintenance of a biological system's functions against internal and external perturbations, mutation, noise, environmental change. Kitano's 2004 review set out the mechanisms, borrowed explicitly from engineering: negative feedback, redundancy, modularity, and decoupling of the phenotype from small variations in the genotype. He also imported the engineer's price, that robustness to common perturbations is bought with fragility to rare ones. Wagner's 2005 monograph made the evolutionary case: neutral networks of genotypes with the same phenotype let mutations accumulate without harm, which is both robustness and the raw material for later change. The function preserved is defined by the organism's fitness, not by a specification.[9][10]
The maintenance of a desired system characteristic despite fluctuations in the behaviour of its components or its environment, Anderies, Janssen, and Ostrom's 2004 definition, taken from engineering and applied to irrigation systems and fisheries. They chose the word deliberately over resilience. Holling's 1973 resilience is the amount of disturbance an ecosystem can absorb before flipping to another state, a property of a system nobody designed. Robustness, in the social-ecological framework, is about institutions people build to hold a system in a state they want, and it carries the engineer's trade-off: robust to the variability the rules anticipate, fragile to the kind they do not. See /resilience for the neighbouring construct.[12][11]
A system's ability to behave acceptably when its inputs or environment violate the assumptions it was written under. Postel's 1980 TCP specification gave the field its slogan (be conservative in what you do, be liberal in what you accept from others) as a rule for implementations facing peers that bend the protocol. Avizienis, Laprie, Randell, and Landwehr's 2004 taxonomy of dependability defines robustness precisely as dependability with respect to external faults, and sets it beside reliability, availability, safety, and integrity; fault tolerance is the means of delivering service despite faults. The faults are discrete events (a malformed packet, a crashed process) and the criterion is the specification.[13][14]
Next
See how robustness sits against neighbouring constructs in the Collider, or export every reference on this page in one click.
Open in ColliderExport BibTeXReport this collisionCited on this page?
If your work is cited behind one of these entries, you know better than we do whether we captured the meaning right. Confirm it, suggest an edit, or dispute it, and your name goes on the entry as an author who has checked it, linked to your ORCID.
Claim your construct