Biased Sampling Strategy Research Articles

Algorithm-Dependent Generalization of AUPRC Optimization: Theory and Algorithm.

Stochastic optimization of the Area Under the Precision-Recall Curve (AUPRC) is a crucial problem for machine learning. Despite extensive studies on AUPRC optimization, generalization is still an open problem. In this work, we present the first trial in the algorithm-dependent generalization of stochastic AUPRC optimization. The obstacles to our destination are three-fold. First, according to the consistency analysis, the majority of existing stochastic estimators are biased with biased sampling strategies. To address this issue, we propose a stochastic estimator with sampling-rate-invariant consistency and reduce the consistency error by estimating the full-batch scores with score memory. Second, standard techniques for algorithm-dependent generalization analysis cannot be directly applied to listwise losses. To fill this gap, we extend the model stability from instance-wise losses to listwise losses. Third, AUPRC optimization involves a compositional optimization problem, which brings complicated computations. In this work, we propose to reduce the computational complexity by matrix spectral decomposition. Based on these techniques, we derive the first algorithm-dependent generalization bound for AUPRC optimization. Motivated by theoretical results, we propose a generalization-induced learning framework, which improves the AUPRC generalization by equivalently increasing the batch size and the number of valid training examples. Practically, experiments on image retrieval and long-tailed classification speak to the effectiveness and soundness of our framework.

Differentially biased sampling strategies reveal the non-stationarity of species distribution models for Indian small felids

Species Distribution Models (SDMs) have been used extensively to understand species-habitat relationships and design conservation strategies. The ability to train these models using a wide variety of datasets and modelling algorithms has led to their wide applicability across systems. However, the ease of modelling also leads to their use as off-the-shelf models without a detailed investigation of the data and their suitable end-use application. The effect of various modelling parameters on inferences has been explored, however, their interaction with training data type is limited. We used country-wide data for four sympatric Indian small cat species to understand the sensitivity of SDMs to data types, sampling extents and their interaction. Our results reveal the non-stationarity of models with varying modelling parameters. The extent of the training dataset had major implications on the inferences and interacted strongly with the type of dataset used. The divergent distribution of the target species revealed that the effect of sampling extent was more pronounced for species that have limited distribution within the predictive extent. Lastly, our results highlight the significance of sampled environmental space in explaining the non-stationarity of the model outputs.

Quantitative Evaluation of the Fracturing State of Crystalline Rocks Using Infrared Thermography

The fracturing state of rocks is a fundamental control on their hydro-mechanical properties. It can be quantified in the laboratory by non-destructive geophysical techniques that are hardly applicable in situ, where biased mapping and statistical sampling strategies are usually exploited. We explore the suitabilty of infrared thermography (IRT) to develop a quantitative, physics-based approach to predict rock fracturing starting from laboratory scales and conditions. To this aim, we performed an experimental study on the cooling behaviour of pre-fractured gneiss and mica schist samples, whose 3D fracture networks were reconstructed using Micro-CT and quantified by unbiased fracture abundance measures. We carried out cooling experiments in both controlled (laboratory) and natural (outdoor) environmental conditions and monitored temperature with a thermal camera. We extracted multi-temporal thermograms to reconstruct the spatial patterns and time histories of temperature during cooling. Their synthetic description show statistically significant correlations with fracture abundance measures. More intensely fractured rocks cool at faster rates and outdoor experiments show that differences in thermal response can be detected even in natural environmental conditions. 3D FEM models reproducing laboratory experiments outline the fundamental control of fracture pattern and convective boundary conditions on cooling dynamics. Based on a lumped capacitance approach, we provided a synthetic description of cooling curves in terms of a Curve Shape Parameter, independent on absolute thermal boundary conditions and lithology. This provides a starting point toward the development of a quantitative methodology for the contactless in situ assessment of rock mass fracturing.

Collision Avoidance Strategy for Unmanned Surface Vessel Considering Actuator Faults Using Kinodynamic Rapidly Exploring Random Tree-Smart and Radial Basis Function Neural Network-Based Model Predictive Control

Path planning and tracking are essential technologies for unmanned surface vessels (USVs). The kinodynamic constraints and actuator faults, however, bring difficulties in finding feasible paths and control efforts. This paper proposes a collision avoidance strategy for USV by developing the kinodynamic rapidly exploring random tree-smart (kinodynamic RRT*-smart) algorithm and the fault-tolerant control method. By utilizing the triangular inequality and the intelligent biased sampling strategy, the kinodynamic RRT*-smart shows its advantages in terms of path length, cost and running time. With consideration of kinodynamic constraints, a feasible and collision-free trajectory can be provided. Then, a radial basis function neural network-based model predictive control (RBF-MPC) method was designed that compensates for the model’s uncertainties by developing the radial basis function neural network (RBF-NN) approximator and by constructing a feedback-state training dataset in real time. Furthermore, two types of fault situation were analyzed considering the thruster failure. We established the faults’ mathematical models and investigated the fault-tolerant strategies for different fault types. The simulation studies were conducted to validate the effectiveness of the proposed strategy. The results show that the proposed planning and control methods can avoid obstacles in faulty conditions.

Biased Sampling and Causal Estimation of Health-Related Information: Laboratory-Based Experimental Research.

BackgroundThe internet is a relevant source of health-related information. The huge amount of information available on the internet forces users to engage in an active process of information selection. Previous research conducted in the field of experimental psychology showed that information selection itself may promote the development of erroneous beliefs, even if the information collected does not.ObjectiveThe aim of this study was to assess the relationship between information searching strategy (ie, which cues are used to guide information retrieval) and causal inferences about health while controlling for the effect of additional information features.MethodsWe adapted a standard laboratory task that has previously been used in research on contingency learning to mimic an information searching situation. Participants (N=193) were asked to gather information to determine whether a fictitious drug caused an allergic reaction. They collected individual pieces of evidence in order to support or reject the causal relationship between the two events by inspecting individual cases in which the drug was or was not used or in which the allergic reaction appeared or not. Thus, one group (cause group, n=105) was allowed to sample information based on the potential cause, whereas a second group (effect group, n=88) was allowed to sample information based on the effect. Although participants could select which medical records they wanted to check—cases in which the medicine was used or not (in the cause group) or cases in which the effect appeared or not (in the effect group)—they all received similar evidence that indicated the absence of a causal link between the drug and the reaction. After observing 40 cases, they estimated the drug–allergic reaction causal relationship.ResultsParticipants used different strategies for collecting information. In some cases, participants displayed a biased sampling strategy compatible with positive testing, that is, they required a high proportion of evidence in which the drug was administered (in the cause group) or in which the allergic reaction appeared (in the effect group). Biased strategies produced an overrepresentation of certain pieces of evidence at the detriment of the representation of others, which was associated with the accuracy of causal inferences. Thus, how the information was collected (sampling strategy) demonstrated a significant effect on causal inferences (F1,185=32.53, P<.001, η2p=0.15) suggesting that inferences of the causal relationship between events are related to how the information is gathered.ConclusionsMistaken beliefs about health may arise from accurate pieces of information partially because of the way in which information is collected. Patient or person autonomy in gathering health information through the internet, for instance, may contribute to the development of false beliefs from accurate pieces of information because search strategies can be biased.

Robustness of network attack strategies against node sampling and link errors.

We investigate the effectiveness of network attack strategies when the attacker has only imperfect information about the network. While most existing network attack strategies assume complete knowledge about the network, in reality it is difficult to obtain the complete structure of a large-scale complex network. This paper considers two scenarios in which the available network information is imperfect. In one scenario, the network contains link errors (i.e., missing and false links) due to measurement errors, and in the other scenario the target network is so large that only part of the network structure is available from network sampling. Through extensive simulations, we show that particularly in a network with highly skewed degree distribution, network attack strategies are robust against link errors. Even if the network contains 30% false links and missing links, the strategies are just as effective as when the complete network is available. We also show that the attack strategies are far less effective when the network is obtained from random sampling, whereas the detrimental effects of network sampling on network attack strategies are small when using biased sampling strategies such as breadth-first search, depth-first search, and sample edge counts. Moreover, the effectiveness of network attack strategies is examined in the context of network immunization, and the implications of the results are discussed.

Combining the biased and unbiased sampling strategy into one convenient free energy calculation method.

Constructing a free energy landscape for a large molecule is difficult. One has to use either a high temperature or a strong driving force to enhance the sampling on the free energy barriers. In this work, we propose a mixed method that combines these two kinds of acceleration strategies into one simulation. First, it applies an adaptive biasing potential to some replicas of the molecule. These replicas are particularly accelerated in a collective variable space. Second, it places some unbiased and exchangeable replicas at various temperature levels. These replicas generate unbiased sampling data in the canonical ensemble. To improve the sampling efficiency, biased replicas transfer their state variables to the unbiased replicas after equilibrium by Monte Carlo trial moves. In comparison to previous integrated methods, it is more convenient for users. It does not need an initial reference biasing potential to guide the sampling of the molecule. And it is also unnecessary to insert many replicas for the requirement of passing the free energy barriers. The free energy calculation is accomplished in a single stage. It samples the data as fast as a biased simulation and it processes the data as simple as an unbiased simulation. The method provides a minimalist approach to the construction of the free energy landscape. © 2019 Wiley Periodicals, Inc.

The $$(1+1)$$ ( 1 + 1 ) Elitist Black-Box Complexity of LeadingOnes

One important goal of black-box complexity theory is the development of complexity models allowing to derive meaningful lower bounds for whole classes of randomized search heuristics. Complementing classical runtime analysis, black-box models help us to understand how algorithmic choices such as the population size, the variation operators, or the selection rules influence the optimization time. One example for such a result is the $$\varOmega (n \log n)$$ lower bound for unary unbiased algorithms on functions with a unique global optimum (Lehre and Witt in Algorithmica 64:623–642, 2012), which tells us that higher arity operators or biased sampling strategies are needed when trying to beat this bound. In lack of analyzing techniques, such non-trivial lower bounds are very rare in the existing literature on black-box optimization and therefore remain to be one of the main challenges in black-box complexity theory. With this paper we contribute to our technical toolbox for lower bound computations by proposing a new type of information-theoretic argument. We regard the permutation- and bit-invariant version of LeadingOnes and prove that its $$(1+1)$$ elitist black-box complexity is $$\varOmega (n^2)$$ , a bound that is matched by $$(1+1)$$ -type evolutionary algorithms. The $$(1+1)$$ elitist complexity of LeadingOnes is thus considerably larger than its unrestricted one, which is known to be of order $$n\log \log n$$ (Afshani et al. in Lecture notes in computer science, vol 8066, pp 1–11. Springer, New York, 2013). The $$\varOmega (n^2)$$ lower bound does not rely on the fact that elitist black-box algorithms are not allowed to make use of absolute fitness values. In contrast, we show that even if absolute fitness values are revealed to the otherwise elitist algorithm, it cannot significantly profit from this additional information. Our result thus shows that for LeadingOnes the memory-restriction, together with the selection requirement, has a substantial impact on the best possible performance.

A simple quantitative method for assessing animal welfare outcomes in terrestrial wildlife shooting: the European rabbit as a case study

AbstractShooting is widely used to reduce the abundances of terrestrial wildlife populations, but there is concern about the animal welfare outcomes (‘humaneness’) of these programmes. Management agencies require methods for assessing the animal welfare outcomes of terrestrial wildlife shooting programmes. We identified four key issues in previous studies assessing the animal welfare outcomes of shooting programmes: (i) biased sampling strategies; (ii) no direct ante mortem observations; (iii) absence of quantifiable parameters for benchmarking; and (iv) no evaluation of explanatory variables that may cause adverse welfare outcomes. We used methods that address these issues to assess the welfare outcomes of a European rabbit (Oryctolagus cuniculus) shooting programme in south-eastern Australia. An independent observer collected ante mortem (distance, timing and outcome of each shot fired) and post mortem (locations of bullet wounds) data. The ante mortem data were used to estimate three critical animal welfare parameters: apparent time to death (ATTD); instantaneous death rate (IDR); and wounding rate (WR). The post mortem data were used to evaluate the location of bullet wounds relative to the Australian national standard operating procedure (SOP). For rabbits, the mean IDR was 0.60, ATTD was 12 s and WR was 0.12. A large proportion of rabbits (0.75) were shot in the cranium or thorax, as required by the SOP. Logistic regression indicated that the proportion of rabbits wounded and missed increased with shooting distance. Hence, reducing shooting distances would increase the humaneness of European rabbit shooting programmes. Our approach enables the animal welfare outcomes of terrestrial shooting programmes to be independently quantified.

Fitting sediment rating curves using regression analysis: a case study of Russian Arctic rivers

Abstract. Published suspended sediment data for Arctic rivers is scarce. Suspended sediment rating curves for three medium to large rivers of the Russian Arctic were obtained using various curve-fitting techniques. Due to the biased sampling strategy, the raw datasets do not exhibit log-normal distribution, which restricts the applicability of a log-transformed linear fit. Non-linear (power) model coefficients were estimated using the Levenberg-Marquardt, Nelder-Mead and Hooke-Jeeves algorithms, all of which generally showed close agreement. A non-linear power model employing the Levenberg-Marquardt parameter evaluation algorithm was identified as an optimal statistical solution of the problem. Long-term annual suspended sediment loads estimated using the non-linear power model are, in general, consistent with previously published results.

Categorization of Multiple Objects in a Scene Using a Biased Sampling Strategy

Recently, various bag-of-features (BoF) methods show their good resistance to within-class variations and occlusions in object categorization. In this paper, we present a novel approach for multi-object categorization within the BoF framework. The approach addresses two issues in BoF related methods simultaneously: how to avoid scene modeling and how to predict labels of an image when multiple categories of objects are co-existing. We employ a biased sampling strategy which combines the bottom-up, biologically inspired saliency information and loose, top-down class prior information for object class modeling. Then this biased sampling component is further integrated with a multi-instance multi-label leaning and classification algorithm. With the proposed biased sampling strategy, we can perform multi-object categorization within an image without semantic segmentation. The experimental results on PASCAL VOC2007 and SUN09 show that the proposed method significantly improves the discriminative ability of BoF methods and achieves good performance in multi-object categorization tasks.

Enhanced representation and multi-task learning for image annotation

In this paper we propose a novel biased random sampling strategy for image representation in Bag-of-Words models. We evaluate its impact on the feature properties and the ranking quality for a set of semantic concepts and show that it improves performance of classifiers in image annotation tasks and increases the correlation between kernels and labels. As second contribution we propose a method called Output Kernel Multi-Task Learning (MTL) to improve ranking performance by transfer information between classes. The main advantages of output kernel MTL are that it permits asymmetric information transfer between tasks and scales to training sets of several thousand images. We give a theoretical interpretation of the method and show that the learned contributions of source tasks to target tasks are semantically consistent. Both strategies are evaluated on the ImageCLEF PhotoAnnotation dataset.Our best visual result which used the MTL method was ranked first according to mean Average Precision (mAP) within the purely visual submissions in the ImageCLEF 2011 PhotoAnnotation Challenge. Our multi-modal submission achieved the first rank by mAP among all submissions in the same competition.

A quorum sensing-mediated switch contributes to natural transformation ofVibrio cholerae

There is a fundamental gap in our understanding of how horizontal gene transfer contributes to the enormous range of genetic variations that are observed among bacteria. The objective of our study was to better understand how the acquisition of genetic material by natural transformation is regulated within a population of Vibrio cholerae cells. V. cholerae is an aquatic bacterium and a facultative human pathogen. It acquires natural competence for transformation in response to changing environmental signals, such as the presence of chitinous surfaces, the absence of monomeric sugars and quorum sensing-linked autoinducers. The latter play a distinctive role in V. cholerae as they fine-tune a switch from the degradation of extracellular DNA toward the uptake of intact DNA strands in competence-induced cells. The link between quorum sensing and natural competence for transformation will be discussed. Furthermore, we speculate on the overrepresentation of transformation-negative strains of V. cholerae in patient-derived culture collections, which might be the result of a biased sampling strategy as virulence and natural transformation are contrarily regulated by the quorum sensing network.

Tabu search for a class of scheduling problems

Scheduling problems are often modeled as resourceconstrained problems in which critical resource assignments to tasks are known and the best assignment of resource time must be made subject to these constraints. Generalization toresource scheduling, where resource assignments are chosen concurrently with times results is a problem which is much more difficult. A simplified model of the general resource scheduling model is possible, however, in which tasks must be assigned a singleprimary resource, subject to constraints resulting from preassignment ofsecondary, or auxiliary, resources. This paper describes extensions and enhancements of tabu search for the special case of the resource scheduling problem described above. The class of problems is further restricted to those where it is reasonable to enumerate both feasible time and primary resource assignments. Potential applications include shift oriented production and manpower scheduling problems as well as course scheduling where classrooms (instructors) are primary and instructors (rooms) and students are secondary resources. The underlying model is a type of quadratic multiple choice problem which we call multiple choice quadratic vertex packing (MCQVP). Results for strategic oscillation and biased candidate sampling strategies are shown for reasonably sized real and randomly generated, synthetic, problem instances. The strategies are compared with other variations using consistent measures of solution time and quality developed for this study.