Search

Robustness is a fundamental pillar of Machine Learning (ML) classifiers, substantially determining their reliability. Methods for assessing classifier robustness are therefore essential. In this work, we address the challenge of evaluating corruption robustness in a way that allows comparability and interpretability on a given dataset. We propose a test data augmentation method that uses a robustness distance $\epsilon$ derived from the datasets minimal class separation distance. The resulting MSCR (mean statistical corruption robustness) metric allows a dataset-specific comparison of different classifiers with respect to their corruption robustness. The MSCR value is interpretable, as it represents the classifiers avoidable loss of accuracy due to statistical corruptions. On 2D and image data, we show that the metric reflects different levels of classifier robustness. Furthermore, we observe unexpected optima in classifiers robust accuracy through training and testing classifiers with different levels of noise. While researchers have frequently reported on a significant tradeoff on accuracy when training robust models, we strengthen the view that a tradeoff between accuracy and corruption robustness is not inherent. Our results indicate that robustness training through simple data augmentation can already slightly improve accuracy., Comment: Accepted for the IJCAI-ECAI-22 Workshop on Artificial Intelligence Safety (AISafety 2022)

Additional file 1.

A set of seasonal drought forecast models was assessed and verified for the Jaguaribe River in semiarid northeastern Brazil. Meteorological seasonal forecasts were provided by the operational forecasting system used at FUNCEME (Ceará's research foundation for meteorology) and by the European Centre for Medium-Range Weather Forecasts (ECMWF). Three downscaling approaches (empirical quantile mapping, extended downscaling and weather pattern classification) were tested and combined with the models in hindcast mode for the period 1981 to 2014. The forecast issue time was January and the forecast period was January to June. Hydrological drought indices were obtained by fitting a multivariate linear regression to observations. In short, it was possible to obtain forecasts for (a) monthly precipitation, (b) meteorological drought indices, and (c) hydrological drought indices.The skill of the forecasting systems was evaluated with regard to root mean square error (RMSE), the Brier skill score (BSS) and the relative operating characteristic skill score (ROCSS). The tested forecasting products showed similar performance in the analyzed metrics. Forecasts of monthly precipitation had little or no skill considering RMSE and mostly no skill with BSS. A similar picture was seen when forecasting meteorological drought indices: low skill regarding RMSE and BSS and significant skill when discriminating hit rate and false alarm rate given by the ROCSS (forecasting drought events of, e.g., SPEI1 showed a ROCSS of around 0.5). Regarding the temporal variation of the forecast skill of the meteorological indices, it was greatest for April, when compared to the remaining months of the rainy season, while the skill of reservoir volume forecasts decreased with lead time.This work showed that a multi-model ensemble can forecast drought events of timescales relevant to water managers in northeastern Brazil with skill. But no or little skill could be found in the forecasts of monthly precipitation or drought indices of lower scales, like SPI1. Both this work and those here revisited showed that major steps forward are needed in forecasting the rainy season in northeastern Brazil.

The Strengths and Difficulties Questionnaire (SDQ) is an internationally popular tool for assessing of psychological and psychosocial development; however, recent research has suggested it lacks measurement invariance in several situations. We expand on this discussion by examining the measurement variance of its two popular models across longitudinal measurements. We examined the fits of both models, conducted tests of invariance across time, and measured the change over time for each factor. We found that both models provide good fits and, furthermore, that strong measurement invariance was upheld across all measurement points for both models. Additionally, we found that problem scores increased and prosocial behavior scores decreased over time. We conclude that both models provide good fits of the data and have measurement invariance over time, but that more research is required regarding its measurement invariance across cultures.

While it is appreciated that global gene expression analyses can provide novel insights about complex biological processes, experiments are generally insufficiently powered to achieve this goal. Here we report the results of a robust microarray experiment of axolotl forelimb regeneration. At each of 20 post-amputation time points, we estimated gene expression for 10 replicate RNA samples that were isolated from 1 mm of heterogeneous tissue collected from the distal limb tip. We show that the limb transcription program diverges progressively with time from the non-injured state, and divergence among time adjacent samples is mostly gradual. However, punctuated episodes of transcription were identified for five intervals of time, with four of these coinciding with well-described stages of limb regeneration-amputation, early bud, late bud, and pallet. The results suggest that regeneration is highly temporally structured and regulated by mechanisms that function within narrow windows of time to coordinate transcription within and across cell types of the regenerating limb. Our results provide an integrative framework for hypothesis generation using this complex and highly informative data set.

Purpose: To examine the physiological and performance responses to a heat-acclimatization camp in highly trained professional team-sport athletes. Methods: Eighteen male Australian Rules Football players trained for 2 wk in hot ambient conditions (3133°C, humidity 3450%). Players performed a laboratory-based heat-response test (24-min walk + 24 min seated; 44°C), a YoYo Intermittent Recovery Level 2 Test (YoYoIR2; indoor, temperate environment, 23°C) and standardized training drills (STD; outdoor, hot environment, 32°C) at the beginning and end of the camp. Results: The heat-response test showed partial heat acclimatization (eg, a decrease in skin temperature, heart rate, and sweat sodium concentration, P < .05). In addition, plasma volume (PV, CO rebreathing, +2.68 [0.83; 4.53] mL/kg) and distance covered during both the YoYoIR2 (+311 [260; 361] m) and the STD (+45.6 [13.9; 77.4] m) increased postcamp (P < .01). None of the performance changes showed clear correlations with PV changes (r < .24), but the improvements in running STD distance in hot environment were correlated with changes in hematocrit during the heat-response test (r = .52, 90%CI [.77; .12]). There was no clear correlation between the performance improvements in temperate and hot ambient conditions (r < .26). Conclusion: Running performance in both hot and temperate environments was improved after a football training camp in hot ambient conditions that stimulated heat acclimatization. However, physiological and performance responses were highly individual, and the absence of correlations between physical-performance improvements in hot and temperate environments suggests that their physiological basis might differ.

In this study we show by means of computational experiments that a pseudo-Boolean approach leads to a very compact presentation of p-Median problem instances which might be solved to optimality by a general purpose solver like CPLEX, Xpress, etc. Together with p-Median benchmark instances from OR and some other libraries we are able to solve to optimality many benchmark instances from cell formation in group technology which were tackled in the past only by means of different types of heuristics. Finally, we show that this approach is flexible to take into account many other practically motivated constraints in cell formation.

In this work, we suggest concepts and solution methodologies for a series of strategic network design problems that find application in highly data-sensitive industries, such as, for instance, the high-tech, governmental, or military sector. Our focus is on the installation of widely used cost-efficient tree-structured communication infrastructure. As base model we use the well-known Steiner tree problem, in which we are given terminal nodes, optional Steiner nodes, and potential network links between nodes. Its objective is to connect all terminals to a distributor node using a tree of minimum total edge costs. The novel, practically relevant side constraints are related to privacy concerns of customers, represented by terminals. In order to account for these, we study four privacy models that restrict the eligible infrastructure for the customer-distributor data exchange: (I) Selected pairs of terminals mutually exclude themselves as intermediate data-transmission nodes; (II) some pairs of terminals require disjoint paths to the distributor; (III) individual terminals forbid routing their data through allegedly untrustworthy links; and (IV) certain terminals do not allow the usage of doubtful links on their entire network branch. These topological data-privacy requirements significantly complicate the notoriously hard optimization problem. We clarify the model relationships by establishing dominance results, point out potential extensions and derive reduction tests. We present corresponding, strong non-compact integer programming (IP) formulations and embed these in efficient cutting plane methods. In addition, we develop constraint programming formulations that are used complementally to derive primal solutions. In a computational study, we analyze the performance of our methods on a diverse set of literature-based test instances.

The paper deals with a sequencing and routing problem originated by a real-world application context. The problem consists in defining the best sequence of locations to visit within a warehouse for the storage and/or retrieval of a given set of items during a specified time horizon, by considering some specific requirements and operating policies which are typical of the kind of warehouse under study. A fleet composed of both electric (i.e., equipped with a lithium-ion battery) and conventional (i.e., with internal combustion engine) forklifts is considered. We model the problem in terms of constrained multicommodity flows on a space-time network, and we extend a matheuristic approach proposed for the case of only conventional vehicles. Preliminary computational results are also presented.

Effective production scheduling allows manufacturing companies to be flexible and well-adjusted to varying customer demand. In practice, production scheduling decisions are subject to several complex constraints which emerge from staff working hours and skills, delivery schedules, stock capacities, machine maintenance and machine setup. This paper introduces a novel production scheduling problem based on the real-world case of a manufacturing company in Belgium. Given a set of customer requests which may only be delivered together on one of the provided potential shipment days, the problem is to select a subset of these requests and schedule the production of the required item quantities subject to the aforementioned restrictions. All decisions must be taken for a time horizon of several days, leading to a complex problem where there may not be enough resources to serve all requests. We provide an integer programming formulation of this novel problem which is capable of solving small-scale instances to proven optimality. In order to efficiently solve large-scale instances, we develop a metaheuristic algorithm. A computational study with instances generated from real-world data indicates that the metaheuristic can quickly produce high-quality solutions, even for cases comprising several days, requests and limited stock capacities. We also conduct a sensitivity analysis concerning characteristics of the schedules and instances, the results of which can be exploited to increase production capacity and revenue.

Cross-docking is a logistics strategy that aims at less transportation costs and fast customer deliveries. Incorporating an efficient vehicle routing could increase the benefits of the cross-docking. In this paper, the vehicle routing problem with reverse cross-docking (VRP-RCD) is studied. Reverse logistics has attracted more attention due to its ability to gain more profit and maintain the competitiveness of a company. VRP-RCD includes a four-level supply chain network: suppliers, cross-dock, customers, and outlets, with the objective of minimizing vehicle operational and transportation costs. A two-phase heuristic that employs an adaptive large neighborhood search (ALNS) with various destroy and repair operators is proposed to solve benchmark instances. The simulated annealing framework is embedded to discover a vast search space during the search process. Experimental results show that our proposed ALNS obtains optimal solutions for 24 out of 30 problems of the first set of benchmark instances while getting better results for all instances in the second set of benchmark instances compared to optimization software. ispartof: pages:167-182 ispartof: COMPUTATIONAL LOGISTICS, ICCL 2020 vol:12433 pages:167-182 ispartof: 11th International Conference on Computational Logistics (ICCL) location:ELECTR NETWORK date:28 Sep - 30 Sep 2020 status: published

Despite the substantial number of existing publications on conceptual modelling education and feedback, in particular, the perfect balance between the effectiveness of feedback and the costs of the feedback design tailored to the field-specific needs of conceptual modelling remains an unanswered scientific and pedagogical question. The existing educational literature and online courses on conceptual modelling tend to overlook the essential aspects of metacognition and self-regulation in the learning process. The problem of providing feedback is exacerbated by the time-consuming nature of manual feedback provision and the difficulties of automating the provision of personalized and elaborated feedback. This paper presents an experience report on designing and implementing a learning ontology-based personalized feedback report aimed at raising student self-awareness and self-regulated learning in a university level conceptual modelling course, while the design aims at automation of the feedback provisioning in the near future. It describes the stages of learning report development and provides directions for adapting this type of feedback for various learning settings in conceptual modelling education, in view of potential future automation of report provision.

This paper addresses the fleet deployment problem and in particular the treatment of inventory in the maritime case. A new model based on time-continuous formulation for the combined maritime fleet deployment and inventory management problem in Roll-on Roll-off shipping is presented. Tests based on realistic data from the Ro-Ro business show that the model yields good solutions to the combined problem within reasonable time. © Springer Verlag. LOCKED until 27.9.2018 due to copyright restrictions. The final publication is available at https://link.springer.com/chapter/10.1007%2F978-3-319-68496-3_22. This is the authors' accepted and refereed manuscript to the article.

At each generation of an ant algorithm, each ant builds a solution step by step by adding an element to it. Each choice is based on the greedy force (short term profit or heuristic information) and the trail system (central memory which collects information during the search process). Usually, all the ants of the population have the same characteristics and behaviors. In contrast in this paper, a new type of ant metaheuristic is proposed. It relies on the use of ants with different personalities. Such a method has been adapted to the well-known vehicle routing problem, and even if it does not match the best known results, its performance is encouraging (on one benchmark instance, new best results have however been found), which opens the door to a new ant algorithm paradigm.