PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Error minimization”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3Linked to original sources
PubMed1999

Algorithmic stability and sanity-check bounds for leave-one-out cross-validation.

In this article we prove sanity-check bounds for the error of the leave-one-out cross-validation estimate of the generalization error: that is, bounds showing that the worst-case error of this estimate is not much worse than that of the training error estimate. The name sanity check refers to the fact that although we often expect the leave-one-out estimate to perform considerably better than the training error estimate, we are here only seeking assurance that its performance will not be considerably worse. Perhaps surprisingly, such assurance has been given only for limited cases in the prior literature on cross-validation. Any nontrivial bound on the error of leave-one-out must rely on some notion of algorithmic stability. Previous bounds relied on the rather strong notion of hypothesis stability, whose application was primarily limited to nearest-neighbor and other local algorithms. Here we introduce the new and weaker notion of error stability and apply it to obtain sanity-check bounds for leave-one-out for other classes of learning algorithms, including training error minimization procedures and Bayesian algorithms. We also provide lower bounds demonstrating the necessity of some form of error stability for proving bounds on the error of the leave-one-out estimate, and the fact that for training error minimization algorithms, in the worst case such bounds must still depend on the Vapnik-Chervonenkis dimension of the hypothesis class.

Algorithms↗
PubMed2000

Early fixation of an optimal genetic code.

The evolutionary forces that produced the canonical genetic code before the last universal ancestor remain obscure. One hypothesis is that the arrangement of amino acid/codon assignments results from selection to minimize the effects of errors (e.g., mistranslation and mutation) on resulting proteins. If amino acid similarity is measured as polarity, the canonical code does indeed outperform most theoretical alternatives. However, this finding does not hold for other amino acid properties, ignores plausible restrictions on possible code structure, and does not address the naturally occurring nonstandard genetic codes. Finally, other analyses have shown that significantly better code structures are possible. Here, we show that if theoretically possible code structures are limited to reflect plausible biological constraints, and amino acid similarity is quantified using empirical data of substitution frequencies, the canonical code is at or very close to a global optimum for error minimization across plausible parameter space. This result is robust to variation in the methods and assumptions of the analysis. Although significantly better codes do exist under some assumptions, they are extremely rare and thus consistent with reports of an adaptive code: previous analyses which suggest otherwise derive from a misleading metric. However, all extant, naturally occurring, secondarily derived, nonstandard genetic codes do appear less adaptive. The arrangement of amino acid assignments to the codons of the standard genetic code appears to be a direct product of natural selection for a system that minimizes the phenotypic impact of genetic error. Potential criticisms of previous analyses appear to be without substance. That known variants of the standard genetic code appear less adaptive suggests that different evolutionary factors predominated before and after fixation of the canonical code. While the evidence for an adaptive code is clear, the process by which the code achieved this optimization requires further attention.

Amino Acids↗
PubMed2001

A statistical method to minimize magnification errors in serial vertebral radiographs.

Incident vertebral deformities are commonly defined by observed changes in height between measurements on two consecutive radiographs. However, conventional radiographs are subject to magnification, and this magnification may differ between films, leading to artifactual changes in height. In order to minimize this effect, it is common practice to record the spine-film and film-focus distances, and from this to calculate a magnification factor for each film. We present a simple statistical method for correcting for differences in magnification between two films if the spine-film and film-focus distances are unknown. This method is shown to reduce the variance of the magnification differences in vertebral heights by 14%, considerably more than is possible using the spine-film distance. Using the statistical method, the number of vertebrae that showed not only a reduction in one or more height of 15%, but were also judged clinically to be free from any incident deformity by an expert radiologist, was reduced from 100 to 46. The number showing a reduction of 20% that were judged fracture-free was reduced from 15 to 9. In the subset of subjects for whom the spine-film distance was known, the reduction in false positives was similar, whichever method was used to correct for magnification. There was no difference in the number of confirmed incident fractures detected when magnification correction by either method was employed. It is concluded that correcting for magnification differences using the statistical method outlined here reduces the number of false positive deformities very substantially and by a similar extent as correcting the magnification using reliable, measured spine-film and film-focus distances. A further advantage of this method is that it can be used retrospectively.

Diagnostic Errors↗
PubMed2001

A protocol for the reduction of systematic patient setup errors with minimal portal imaging workload.

PURPOSE: To evaluate a new off-line patient setup correction protocol that minimizes the required number of portal images and perform a comparison with currently applied protocols. METHODS AND MATERIALS: We compared two types of off-line protocols: (a) the widely applied shrinking action level (SAL) protocol, in which the setup error, averaged over the measured treatment fractions, is compared with a threshold that decreases with the number of measurements, to decide if a correction is necessary; and (b) a new "no-action-level" (NAL) protocol, which simply calculates the mean setup error over a fixed number of fractions, and always corrects for it. The performance of the protocols was evaluated by applying them to (a) a database of measured setup errors from 600 prostate patients (with, on average, 10 imaged fractions/patient) and (b) Monte Carlo-generated setup error distributions for various values of the population systematic and random errors. RESULTS: The NAL protocol achieved a significantly higher accuracy than the SAL protocol for a similar workload in terms of image acquisition and analysis, as well as in setup corrections. The SAL protocol required approximately three times more images than the NAL protocol to obtain the same reduction of systematic errors. Application of the NAL protocol to measured setup errors confirmed its efficacy in systematic error reduction in a real patient population. CONCLUSION: The NAL protocol performed much more efficiently than the SAL protocol for both actually measured and simulated setup data. The resulting decrease in required portal images not only reduces workload, but also dose to healthy tissue, if dedicated large fields are required for portal imaging (double exposure).

Algorithms↗
PubMed2000

Cytologic features of proliferative breast disease: a study designed to minimize sampling error.

BACKGROUND: Assessment of cytologic features that allow accurate classification of proliferative breast disease has been hampered by sampling errors when fine-needle aspirations have been compared with their corresponding histologic sections. METHODS: To allow for optimal cytohistologic correlation, 2 smears (1 hematoxylin and eosin-stained and 1 Diff-Quik-stained) were prepared from each of 98 breast biopsies without mass lesions and compared with the corresponding histologic sections of the scraped area. Each smear was reviewed in a blinded fashion and assessed for cellularity, background elements, cytoarchitectural features of cell groups, and nuclear features by 2 reviewers. Smears were then classified as nonproliferative breast disease (NPBD), proliferative breast disease without atypia (PBD) or with atypia (PBDA), or DCIS, based on review of the corresponding histologic sections. RESULTS: When comparing NPBD/PBD (n = 86) with PBDA/DCIS (n = 12), smears from PBDA/DCIS were significantly (by the Fisher exact test or Wilcoxon rank sum P values with adjustment for multiple comparisons) more likely to be cellular; contain single cells and necrosis; exhibit nuclear overlap and cytoplasmic vacuoles; have large nuclei, macronucleoli, pleomorphism, clumped chromatin, and hyperchromasia; and were less likely to have complex cell groups, monolayers, swirling, cohesion, and myoepithelial cells in epithelial sheets and the smear background. When NPBD (n = 53) and PBD (n = 33) were similarly compared, smears from PBD were more likely to exhibit larger and more complex cell groups, but they were otherwise similar to smears from NPBD. CONCLUSIONS: There are many cytologic features that will allow a distinction of NPBD/PBD from PBDA/DCIS, but relatively few that can aid in separating NPBD from PBD.

Biopsy↗
PubMed1976

Minimizing biopsy error.

Several factors influence substantially the order of accuracy of oral biopsy procedures. This paper evaluates some of these factors and suggests methods whereby accuracy can be improved. The surgeon's responsibility extends beyond the operating theater and into the laboratory if maximum value is to be derived from each biopsy specimen. It is suggested that present sucess rates of surgical treatment of oral carcinoma can be improved by attention to laboratory procedures by the clinician.

Biopsy↗
PubMed1981

Minimizing sample error in transmission electron microscopy.

The last few years have seen an increase in the use of electron microscopy, both for research and as a diagnostic tool in pathology. With the increase in the number of specimens being used for diagnostic purposes, limitations due to sampling error have become apparent. The method described is designed to help overcome this problem.

Microscopy, Electron↗
PubMed2000

Minimizing binding errors using learned conjunctive features.

We have studied some of the design trade-offs governing visual representations based on spatially invariant conjunctive feature detectors, with an emphasis on the susceptibility of such systems to false-positive recognition errors-Malsburg's classical binding problem. We begin by deriving an analytical model that makes explicit how recognition performance is affected by the number of objects that must be distinguished, the number of features included in the representation, the complexity of individual objects, and the clutter load, that is, the amount of visual material in the field of view in which multiple objects must be simultaneously recognized, independent of pose, and without explicit segmentation. Using the domain of text to model object recognition in cluttered scenes, we show that with corrections for the nonuniform probability and nonindependence of text features, the analytical model achieves good fits to measured recognition rates in simulations involving a wide range of clutter loads, word size, and feature counts. We then introduce a greedy algorithm for feature learning, derived from the analytical model, which grows a representation by choosing those conjunctive features that are most likely to distinguish objects from the cluttered backgrounds in which they are embedded. We show that the representations produced by this algorithm are compact, decorrelated, and heavily weighted toward features of low conjunctive order. Our results provide a more quantitative basis for understanding when spatially invariant conjunctive features can support unambiguous perception in multiobject scenes, and lead to several insights regarding the properties of visual representations optimized for specific recognition tasks.

Artificial Intelligence↗
PubMed2000

Minimizing binding errors using learned conjunctive features.

We have studied some of the design trade-offs governing visual representations based on spatially invariant conjunctive feature detectors, with an emphasis on the susceptibility of such systems to false-positive recognition errors-Malsburg's classical binding problem. We begin by deriving an analytical model that makes explicit how recognition performance is affected by the number of objects that must be distinguished, the number of features included in the representation, the complexity of individual objects, and the clutter load, that is, the amount of visual material in the field of view in which multiple objects must be simultaneously recognized, independent of pose, and without explicit segmentation. Using the domain of text to model object recognition in cluttered scenes, we show that with corrections for the nonuniform probability and nonindependence of text features, the analytical model achieves good fits to measured recognition rates in simulations involving a wide range of clutter loads, word size, and feature counts. We then introduce a greedy algorithm for feature learning, derived from the analytical model, which grows a representation by choosing those conjunctive features that are most likely to distinguish objects from the cluttered backgrounds in which they are embedded. We show that the representations produced by this algorithm are compact, decorrelated, and heavily weighted toward features of low conjunctive order. Our results provide a more quantitative basis for understanding when spatially invariant conjunctive features can support unambiguous perception in multiobject scenes, and lead to several insights regarding the properties of visual representations optimized for specific recognition tasks.

Algorithms↗
PubMed1985

Goniomegaly associated with a normal cornea, increased axial length, and minimal refractive error.

The association of a large axial length with a small refractive error and a normal corneal diameter should alert the lens implant surgeon to the possibility that goniomegaly is present. We report a patient whose intraoperative anterior chamber diameter was measured as 15 mm. It was necessary to insert a modified posterior chamber intraocular lens instead of an anterior chamber lens.

Anterior Chamber↗