Selection Of Test Samples Research Articles

Uncertainty guided test-time training for face forgery detection

The rapid development of generative image modeling poses security risks of spreading unreal visual information, even though those techniques make a lot of applications possible in positive aspects. To provide alerts and maintain a secure social environment, forgery detection has been an urgent and crucial solution to deal with this situation and try to avoid any negative effects, especially for human faces, owing to potential severe results when malicious creators spread disinformation widely. In spite of the success of recent works w.r.t. model design and feature engineering, detecting face forgery from novel image creation methods or data distributions remains unresolved, because well-trained models are typically not robust to the distribution shift during test-time. In this work, we aim to alleviate the sensitivity of an existing face forgery detector to new domains, and then boost real-world detection under unknown test situations. In specific, we leverage test examples, selected by uncertainty values, to fine-tune the model before making a final prediction. Therefore, it leads to a test-time training based approach for face forgery detection, that our framework incorporates an uncertainty-driven test sample selection with self-training to adapt a classifier onto target domains. To demonstrate the effectiveness of our framework and compare with previous methods, we conduct extensive experiments on public datasets, including FaceForensics++, Celeb-DF-v2, ForgeryNet and DFDC. Our results clearly show that the proposed framework successfully improves many state-of-the-art methods in terms of better overall performance as well as stronger robustness to novel data distributions.

HashC: Making deep learning coverage testing finer and faster

Though Deep Neural Networks (DNNs) have been widely deployed and achieved great success in many domains, they have severe safety and reliability concerns. To provide testing evidence for DNNs’ reliable behaviors, various coverage testing techniques inspired by traditional software testing have been proposed. However, the coverage criteria in these techniques are either not fine enough to capture subtle behaviors of DNNs, or too time-consuming to be applied on large-scale DNNs. In this paper, we propose a coverage testing framework named HashC, which makes mainstream coverage criteria (e.g., NC and KMNC) much finer. Meanwhile, HashC reduces the time complexity of combinatorial coverage testing from polynomial time to linear time. We also develop the corresponding test sample selection method. Our experiments show that, (1) the existing mainstream coverage criteria are becoming finer after being equipped with HashC, (2) HashC greatly accelerates combinatorial coverage testing and can handle the testing of large-scale DNNs.

A Review of Pattern Recognition Using Statistics

Abstract: Pattern recognition's primary objective is supervised or classification techniques. The statistical method has been the most intensively researched and implemented in practice among the various frameworks in which classification technique has typically been developed. In recent times, neural network techniques and methods based on statistical learning theory have gotten a lot of attention. The following issues must be carefully considered when designing a recognition system: pattern classes, sensing environment, pattern representation, feature extraction and selection, cluster analysis, classifier design, learning, training, test sample selection, and performance evaluation. The overall challenge of detecting intricate forms with modifiable direction, position, and size has not been addressed despite just Fifty years of intensive research. Data mining, web browsing, recovery of multimedia content, face recognition, and cursive handwriting recognition are just a couple of minor new and emerging applications that require strong and fast pattern recognition methods. The intention of this review paper is to summarize and analyze some of the adopted approaches in various phases of a recognition system, as well as to identify areas of research and applications that are at the forefront of innovation in this exciting and challenging field.

Surface Defect Classification in Silicon Wafer Manufacturing Using the Linear-Based Channeling and Rule-Based Binning Algorithms

Developing an accurate means of classifying defects, such as crystal-originated pits, surface-adhered foreign particles, and process-induced defects, using scanning surface inspection systems (SSIS) is of paramount importance because it provides the opportunity to determine the root causes of defects, which is valuable for yield enhancement. This report presents a novel defect classification approach developed by optimizing the linear-based channeling (LBC) and rule-based binning (RBB) algorithms that are applied to a commercially available SSIS (KLA-SP5), in combination with test sample selection including the signature defect patterns associated with the typical crystal growth process. The experimental results demonstrate that defect classification is possible with an accuracy and purity above 80% using the LBC algorithm and 90% using the RBB algorithm.

Model and application of farmers' credit risk early warning system based on T-S fuzzy neural network application.

In China, farmers' loan difficulties have become a major problem restricting increases in farmers' incomes and the economic development of rural areas. The existing studies of the management and control of farmers' credit risk have mostly been pre-management, which cannot efficiently prevent and reduce the occurrence of farmers' credit risk in time. This paper uses the T-S neural network model to build a farmers' credit risk early warning system so that formal financial institutions can predict the occurrence of and changes in the farmers' credit risks in a timely manner and quickly undertake countermeasures to reduce losses. After training and testing, a model with a higher degree of fit is used to analyze the credit level of farmers in Shaanxi Province from 2016 to 2018. The results demonstrate that the credit level of farmers in this area is continuously improving, in agreement with the actual situation. The results also show that the prediction accuracy of the T-S fuzzy neural network is high, verifying the rationality of the selection of test samples.

Clinical applicability of automated cephalometric landmark identification: Part I-Patient-related identification errors.

To determine whether AI systems that recognize cephalometric landmarks can apply to various patient groups and to examine the patient-related factors associated with identification errors. The present retrospective cohort study analysed digital lateral cephalograms obtained from 1785 Japanese orthodontic patients. Patients were categorized into eight subgroups according to dental age, cleft lip and/or palate, orthodontic appliance use and overjet. An AI system that automatically recognizes anatomic landmarks on lateral cephalograms was used. Thirty cephalograms in each subgroup were randomly selected and used to test the system's performance. The remaining cephalograms were used for system learning. The success rates in landmark recognition were evaluated using confidence ellipses with α=0.99 for each landmark. The selection of test samples, learning of the system and evaluation of the system were repeated five times for each subgroup. The mean success rate and identification error were calculated. Factors associated with identification errors were examined using a multiple linear regression model. The success rate and error varied among subgroups, ranging from 85% to 91% and 1.32mm to 1.50mm, respectively. Cleft lip and/or palate was found to be a factor associated with greater identification errors, whereas dental age, orthodontic appliances and overjet were not significant factors (all, P<.05). Artificial intelligence systems that recognize cephalometric landmarks could be applied to various patient groups. Patient-oriented errors were found in patients with cleft lip and/or palate.

Deciphering Cancer Cell Behavior From Motility and Shape Features: Peer Prediction and Dynamic Selection to Support Cancer Diagnosis and Therapy.

Cell motility varies according to intrinsic features and microenvironmental stimuli, being a signature of underlying biological phenomena. The heterogeneity in cell response, due to multilevel cell diversity especially relevant in cancer, poses a challenge in identifying the biological scenario from cell trajectories. We propose here a novel peer prediction strategy among cell trajectories, deciphering cell state (tumor vs. nontumor), tumor stage, and response to the anticancer drug etoposide, based on morphology and motility features, solving the strong heterogeneity of individual cell properties. The proposed approach first barcodes cell trajectories, then automatically selects the good ones for optimal model construction (good teacher and test sample selection), and finally extracts a collective response from the heterogeneous populations via cooperative learning approaches, discriminating with high accuracy prostate noncancer vs. cancer cells of high vs. low malignancy. Comparison with standard classification methods validates our approach, which therefore represents a promising tool for addressing clinically relevant issues in cancer diagnosis and therapy, e.g., detection of potentially metastatic cells and anticancer drug screening.

Towards Improved Design and Evaluation of Epileptic Seizure Predictors.

Key issues in the epilepsy seizure prediction research are (1) the reproducibility of results (2) the inability to compare multiple approaches directly. To overcome these problems, the seizure prediction challenge was organized on Kaggle.com. It aimed at establishing benchmarks on a dataset with predefined train, validation, and test sets. Our main objective is to analyze the competition format, and to propose improvements, which would facilitate a better comparison of algorithms. The second objective is to present a novel deep learning approach to seizure prediction and compare it to other commonly used methods using patient centered metrics. We used the competition's datasets to illustrate the effects of data contamination. Having better data partitions, we compared three types of models in terms of different objectives. We found that correct selection of test samples is crucial when evaluating the performance of seizure forecasting models. Moreover, we showed that models, which achieve state-of-the-art performance with respect to commonly used AUC, sensitivity, and specificity metrics, may not yet be suitable for practical usage because of low precision scores. Correlation between validation and test datasets used in the competition limited its scientific value. Our findings provide guidelines which allow for a more objective evaluation of seizure prediction models.

Willingness to pay for improved energy: Evidence from Kenya

Using a sample respondents of 3665 households and 1669 enterprises, this paper utilizes a double bound, open-ended, contingent valuation approach where those who respond positively to the willingness to pay question, were asked to state the maximum amount they were willing to pay as quality levy. The paper identifies protest responses as those respondents who are not willing to pay for the proposed program or as outliers who may state a willingness to pay value either higher or lower than the average willingness to pay value. Bidders and genuine zeros responses are respondents that either state a zero or a positive willingness to pay respectively. Heckman’s sample-selection procedure is used to test sample selection bias in and also analyze the WTP function.The mean household willingness to pay as quality levy was estimated to be US$ 6.53, US$ 3.85 and US $ 6.34 monthly for firewood, charcoal and electricity respectively while the enterprise willingness to pay as quality levy was estimated to be US $ 355.92 monthly. Income, type of dwelling, education, gender, price, location, type of enterprise and size of employees were identified as important factors that explained the differences in the WTP variations.

Moisture uptake and permeability of canvas paintings and their components

Abstract Canvas paintings may show significant dimensional changes and experience internal stresses with fluctuating relative humidity. The relatively high and rapid absorption and drying of moisture within the different layers makes them more vulnerable than panel or wall paintings in comparable conditions. The dynamics of the moisture response is controlled by the water vapour permeability of the different layers. This paper presents a quantitative investigation of the vapour sorption and permeability of a selection of canvas painting components and of reconstructed paintings made of them. The selection of test samples was based on a survey of the materials used by Cuno Amiet in his early work and encompasses linen canvas, collagen glue sizing, chalk-glue ground and brown umber pigmented oil paint. Dynamic Vapour Sorption (DVS) tests were performed to obtain sorption isotherms. The vapour permeability was analysed in terms of the vapour resistance of layers and measured by means of wet cup and dry cup tests as well as in double chamber tests. The principle of incremental resistances was used to discriminate between the properties of the different layers. Whereas glue and canvas are comparable in being strongly absorbent, it appears that their vapour resistance is very different: a continuous glue film has a much higher vapour resistance than a canvas. In this context, we found that the method of applying glue sizing on a canvas influences the permeability of the resulting sized canvas: a gel size forms a more continuous glue film and hence leads to higher vapour resistance of the system, as opposed to a liquid size. Chalk-glue grounds have low moisture sorption, when compared to the high absorption of the proteinaceous glue, because they consist largely of chalk particles, which are not hygroscopic. The umber oil paint stands out for its low sorption and its high resistance to vapour transfer. These results characterise the highly heterogeneous nature of the multi-layered system of a painting in a quantitative way, enabling to better interpret damage phenomena and to make computational predictions of the influence of changing boundary conditions.

Toward a Replacement of the CIE Color Rendering Index for White Light Sources

ABSTRACTThe CIE color rendering (fidelity) index (CRI) has remained unchanged for over four decades. Most, if not all, of its components could be updated to more state-of-the-art methods. One of the most critical components of any color rendering (fidelity) metric is the test sample selection. This article therefore addresses the importance of uniform sampling of wavelength space to avoid selective optimization—that is, taking advantage of the unequal contributions of different wavelength regions to the general color rendering score—of light source spectral power distributions. It summarizes the development of a mathematical sample set with undistorted spectral sensitivity, the HL17 set. The set is used in a recently proposed update, the CRI2012 general color rendering index. To assess the impact of the spectrally uniform sample set on color fidelity scores, the CRI2012 index values for each of a set of 139 lamps were compared with those of the CIE CRI. In addition, the impact of updating the other components was investigated. A mean and maximum absolute difference of respectively 5.9 and 21.8 index units were found between the CRI2012 and CIE CRI, although the largest part—respectively 4.03 and 19.7 index units—was shown to be the result of updating the color difference engine and the switch to the CIE 10° observer. The analysis also indicated possible past spectral selective optimization of some warm-white tri-band fluorescent sources for high luminous efficacy of radiation (LER) and (just) sufficient CIE Ra values by taking advantage of the spectral nonuniformity of the CIE reflectance set. Adopting a spectrally uniform sample set in a color rendering metric therefore has important practical implications when designing light source spectra. Finally, possible updates and further improvements of the CRI2012 are briefly mentioned.

Verification of an optical micro-CMM using the focus variation technique: Aspects of probing errors

Optical systems based on the focus variation technique are able to collect surface topography measurements from machined parts at high speed and a considerably lower cost than a contacting coordinate measuring machine (CMM). This paper presents a preliminary study of probing characteristics of a focus variation instrument within a single field of view, following the guidance of the international standard ISO 10360-8 (a recently published standard for acceptance and reverification testing of CMMs which use optical distance sensors). The selection of test samples and test strategies is also discussed, with specific emphasis on the adaptations made to the ISO 10360-8 guidance to realise accurate measurements.

Colour fidelity for picture gallery illumination, Part 2: Test sample selection – museum tests

An earlier paper has described a new method to optimise the spectral power distribution of solid state lighting systems used in museums. This paper will discuss the question of the selection of the test samples used for the spectral power distribution optimisation for the illumination of renaissance paintings. For fine tuning the spectrum the help of museum curators was needed, who remember the appearance of the pictures under traditional light sources and who were able to advise lighting engineers to tune the spectrum to obtain a pleasing appearance. The paper will show the results obtained by testing the optimum spectrum in the museum. Part 1 of the present publication dealt with a new method to optimise the spectral power distribution of Solid State Lighting systems used in museums. Part 2 will discuss the question of test samples for the illumination of renaissance paintings and will show results obtained by testing the optimum spectrum in the museum.

The Architect Test Sample Selection Model and its Application Based on Genetic Algorithm

This paper introduces the researches and applications based on genetic algorithm in the sample selection of papers. We followed the steps of the work and separate the model of four sub models: the pre-process model, sample selection model, property model and reefed model. We consider form all aspect which may affects the decision of sample selection and solve them. Form the instance, the genetic algorithm we discussed can improve the reliability and efficiency of the national first degree architecture test on some degree.

Tumor heterogeneity: mechanisms and bases for a reliable application of molecular marker design.

Tumor heterogeneity is a confusing finding in the assessment of neoplasms, potentially resulting in inaccurate diagnostic, prognostic and predictive tests. This tumor heterogeneity is not always a random and unpredictable phenomenon, whose knowledge helps designing better tests. The biologic reasons for this intratumoral heterogeneity would then be important to understand both the natural history of neoplasms and the selection of test samples for reliable analysis. The main factors contributing to intratumoral heterogeneity inducing gene abnormalities or modifying its expression include: the gradient ischemic level within neoplasms, the action of tumor microenvironment (bidirectional interaction between tumor cells and stroma), mechanisms of intercellular transference of genetic information (exosomes), and differential mechanisms of sequence-independent modifications of genetic material and proteins. The intratumoral heterogeneity is at the origin of tumor progression and it is also the byproduct of the selection process during progression. Any analysis of heterogeneity mechanisms must be integrated within the process of segregation of genetic changes in tumor cells during the clonal expansion and progression of neoplasms. The evaluation of these mechanisms must also consider the redundancy and pleiotropism of molecular pathways, for which appropriate surrogate markers would support the presence or not of heterogeneous genetics and the main mechanisms responsible. This knowledge would constitute a solid scientific background for future therapeutic planning.

Test sample selection by preschool children: Honoring diversity

In a test of an inductive inference, preschool children's selection of objects was examined. In Experiment 1, 4-year-olds selected diverse objects first in sequential selections; in Experiment 2, adults and 4-year-olds, but not 3-year-olds, made similar selections under the same conditions. A defective object led subjects in all age groups to test a similar object. In Experiment 3, 4-year-olds chose to test a pair of dissimilar objects rather than a pair of similar objects, but 3-year-olds did not. Three-year-olds' selections were independent of diversity. In Experiment 4, we attempted to emphasize the diversity of objects for 3-year-olds. Their first task was to select an object that was the same as or different from a target object. The subjects responded correctly in this task but did not prefer to test diverse objects. Experiment 5 showed that neither 3- nor 4-year-olds have a bias to select nondiverse objects in a nontest context. The findings indicate that children as young as 4 years old value diverse evidence in induction.

Making One Measurement do the Work of Two

Frequently measurements are made under conditions which are either hard to specify precisely or difficult to hold constant for any considerable period. Corrections for drifts or shifts arising from these uncontrolled conditions are often based on measurements made upon control or standard samples periodically introduced in the work schedule. These standard samples make possible the adjustment of the measurements on the test samples at the price of diverting effort that might otherwise be spent on test samples. The standard samples may be dispensed with by picking out certain ones of the test samples for measurement at a later time. This paper presents some schedules for the selection of test samples for remeasurement. When the schedule possesses a balanced symmetry the arithmetical operations for adjusting the observations become simple and easy. Furthermore, all the measurements made contribute information on the test sample.