The evolution of continuous learning of the structure of the environment

Continuous, ‘always on’, learning of structure from a stream of data is studied mainly in the fields of machine learning or language acquisition, but its evolutionary roots may go back to the first organisms that were internally motivated to learn and represent their environment. Here, we study under what conditions such continuous learning (CL) may be more adaptive than simple reinforcement learning and examine how it could have evolved from the same basic associative elements. We use agent-based computer simulations to compare three learning strategies: simple reinforcement learning; reinforcement learning with chaining (RL-chain) and CL that applies the same associative mechanisms used by the other strategies, but also seeks statistical regularities in the relations among all items in the environment, regardless of the initial association with food. We show that a sufficiently structured environment favours the evolution of both RL-chain and CL and that CL outperforms the other strategies when food is relatively rare and the time for learning is limited. This advantage of internally motivated CL stems from its ability to capture statistical patterns in the environment even before they are associated with food, at which point they immediately become useful for planning.     

Geographic profiling applied to testing models of bumble-bee foraging

Geographic profiling (GP) was originally developed as a statistical tool to help police forces prioritize lists of suspects in investigations of serial crimes. GP uses the location of related crime sites to make inferences about where the offender is most likely to live, and has been extremely successful in criminology. Here, we show how GP is applicable to experimental studies of animal foraging, using the bumble-bee Bombus terrestris. GP techniques enable us to simplify complex patterns of spatial data down to a small number of parameters (2–3) for rigorous hypothesis testing. Combining computer model simulations and experimental observation of foraging bumble-bees, we demonstrate that GP can be used to discriminate between foraging patterns resulting from (i) different hypothetical foraging algorithms and (ii) different food item (flower) densities. We also demonstrate that combining experimental and simulated data can be used to elucidate animal foraging strategies: specifically that the foraging patterns of real bumble-bees can be reliably discriminated from three out of nine hypothetical foraging algorithms. We suggest that experimental systems, like foraging bees, could be used to test and refine GP model predictions, and that GP offers a useful technique to analyse spatial animal behaviour data in both the laboratory and field.     

Heuristic approach for assigning workers to tasks based on individual learning rates

A heuristic worker-task assignment based on individual worker learning rates is examined for two tasks, one with a long production run, the other with a short production run. Simulations of total productivity are performed under several experimental conditions based on empirical industrial worker productivity measures. Results indicate that the heuristic method significantly improves overall productivity under empirically observed conditions and under many experimental conditions. The heuristic policy performs best in conditions where workers learn more gradually. Conditions are discussed where the heuristic policy provides the greatest potential for improvement based on factors including the mixture of production cycles, mean learning rate, mean forgetting rate, mean prior expertise, variance of prior expertise and variance of steady-state productivity.     

Kirkpatrick's Level 1 Evaluation of the Implementation of a Computer‐Aided Process Design Tool in a Senior‐Level Engineering Course

Computer simulation tools are frequently used in engineering design work, and undergraduates are often trained to use these tools as they learn to design systems. The use of new tools in the learning environment should be evaluated to assure that the students are able to use the tools effectively. This study details and demonstrates the use of a Kirkpatrick's Level 1 Evaluation to assess the effectiveness of an instructional environment in which students learn to use a computer simulation tool to perform engineering design work. Specifically, an evaluation was conducted to look at student perceptions of FOODS‐LIB—a steady‐state food process design tool, its user's manual learning modules, and the implementation of FOODS‐LIB in a senior level design course. This evaluation was triangulated with an instructor's assessment of student products generated as the students used the learning modules and designed an ice cream manufacturing process using FOODS‐LIB.     

Promoting Active Learning with Cases and Instructional Modules

In this paper, we propose the use of cases and instructional modules to teach invention, engineering design, and elements of technology management. One way to learn is to study and reflect upon the experience of others. Such experience may be captured in a case. Cases promote active learning by requiring students to assume the roles of participants in the decision making process. Cases are also a vehicle for raising business issues and human resources concerns not usually considered in traditional engineering courses. Real world design and engineering involves risk and uncertainty, tradeoffs and priorities, ethical issues, human elements, and impact assessment. Cases expose students to open ended, ill defined problems whose solution often depends on making assessments, judgments, and decisions about the technical competencies of the organization.     

An exploratory study of risk aversion in supply chain dynamics via human experiment and agent-based simulation

The literature on the impact of risk aversion on supply chains (SCs) is relatively limited and, in particular, there is a dearth of theory and a lack of empirical evidence concerning: (1) the impact of individual risk aversion on the generation and dynamics of the order policy (e.g. order patterns and inventory holding costs); (2) the impact of several combinations of risk-averse members in each stage of a multi-echelon SC. We explore these gaps through a multi-method approach (i.e. human experiments and agent-based simulation), thus using both empirical and simulated data. Specifically, based on results from a human experiment, we develop the conjecture that risk aversion is positively correlated to the desired stock level and consequently to the safety stock factor of inventory order policies. Building on this finding, we perform a simulation study to infer the impact of individual risk aversion in a multi-echelon SC. Results show that alternative compositions of the SC in terms of risk aversion levels of the echelons significantly influence inventory holdings and SC dynamics. The study implies that a company facing problems of high inventory days-on-hand should favour low-risk aversion managers, as instrumental to lowering stock and improving net working capital.     

Attentional control, high intensity pleasure, and risky pedestrian behavior in college students

Individual differences in temperament and personality are closely linked to motor vehicle safety. However, 13% of Americans who die in transportation-related injuries are not killed in motor vehicle crashes, but rather in pedestrian injuries. This study was designed to study links between two individual difference measures, attentional control and high intensity pleasure, and pedestrian injury risk among college students, a group at particular risk of pedestrian injury. A sample of 245 students completed a temperament questionnaire and engaged in a street-crossing task within an interactive, immersive virtual pedestrian environment. Individuals scoring high on attentional control (the capacity to focus and shift attention, one facet of conscientiousness) waited longer to choose gaps to cross within and showed some tendency to choose larger gaps after waiting. Individuals scoring high in high intensity pleasure (the tendency to desire novel, complex, and varied stimuli, one facet of sensation-seeking) were more likely to experience collisions with traffic in the virtual environment. Theoretical and applied implications are discussed.     

风险规避对制造商开通直销渠道的影响

研究了风险规避对制造商开通直销渠道的影响。运用博弈论中个人理性和激励相容原则建立单渠道和双渠道供应链需求模型,得出均衡情况下供应链各成员的最优决策及社会福利。研究发现,仅当决策者风险规避度和直销渠道销售成本满足一定条件时,制造商开通直销渠道有利可图;且直销渠道并非总侵蚀零售商效用。当决策者风险规避度较大且直销渠道成本适中时,供应链价值减少。社会福利总随着制造商风险规避度的增大呈现先减少后增加的趋势。     

Scholarly communication as a socioecological system

Among Belver C. Griffith's many contributions to disciplinary communication is the idea that science and scholarship at large constitute a social system to be investigated empirically. This paper reports findings of an author co-citation analysis of the field of human behavioral ecology that expands Griffith's concept of the social system of scientific communication to fit a socioecological framework. Cluster analysis and multidimensional scaling techniques are used to characterize the research specialty at large and portray five respondents' individual resource maps. The techniques reveal co-citation relationships among authors whose work they had referenced in recent articles. Survey data on searching and handling behaviors for an aggregated sample of 180 cited references are correlated with core-periphery zones of the individual maps. Findings that types of socially mediated communication and distinctive information foraging behaviors correlate with different zones of a bibliographic microhabitat support an interpretation that active specialty members conform to foraging efficiency principles as predicted by prey-choice models from optimal foraging theory. This revised version was published online in June 2006 with corrections to the Cover Date.     

A Novel Combinational Convolutional Neural Network for Automatic Food-Ingredient Classification

With the development of deep learning and Convolutional Neural Networks (CNNs), the accuracy of automatic food recognition based on visual data have significantly improved. Some research studies have shown that the deeper the model is, the higher the accuracy is. However, very deep neural networks would be affected by the overfitting problem and also consume huge computing resources. In this paper, a new classification scheme is proposed for automatic food-ingredient recognition based on deep learning. We construct an up-to-date combinational convolutional neural network (CBNet) with a subnet merging technique. Firstly, two different neural networks are utilized for learning interested features. Then, a well-designed feature fusion component aggregates the features from subnetworks, further extracting richer and more precise features for image classification. In order to learn more complementary features, the corresponding fusion strategies are also proposed, including auxiliary classifiers and hyperparameters setting. Finally, CBNet based on the well-known VGGNet, ResNet and DenseNet is evaluated on a dataset including 41 major categories of food ingredients and 100 images for each category. Theoretical analysis and experimental results demonstrate that CBNet achieves promising accuracy for multi-class classification and improves the performance of convolutional neural networks.     

Predictive ethoinformatics reveals the complex migratory behaviour of a pelagic seabird,the Manx Shearwater

Understanding the behaviour of animals in the wild is fundamental to conservation efforts. Advances in bio-logging technologies have offered insights into the behaviour of animals during foraging, migration and social interaction. However, broader application of these systems has been limited by device mass, cost and longevity. Here, we use information from multiple logger types to predict individual behaviour in a highly pelagic, migratory seabird, the Manx Shearwater (Puffinus puffinus). Using behavioural states resolved from GPS tracking of foraging during the breeding season, we demonstrate that individual behaviours can be accurately predicted during multi-year migrations from low cost, lightweight, salt-water immersion devices. This reveals a complex pattern of migratory stopovers: some involving high proportions of foraging, and others of rest behaviour. We use this technique to examine three consecutive years of global migrations, revealing the prominence of foraging behaviour during migration and the importance of highly productive waters during migratory stopover.     

Impact of Holistic and Learning‐Oriented Teaching on Academic Success

The influence of the preferred learning styles and psychological types on academic performance have been assessed for engineering students. The subjects of the experiments were different groups of students taking the first and the second courses in the civil engineering department. A holistic teaching/learning approach was designed based on the fact that people have different psychological characters and learn in different ways. During the course, many teaching/learning techniques were introduced to foster the learning process for different learning styles and psychological types. The study explores the implications of the learning styles and the psychological temperaments on the academic performance for experimental and control groups. The Learning Type Measure (LTM) and the Myers Briggs Type Indicator (MBTI) instruments were used to assess the psychological profiles and the learning preferences of the students. The paper will discuss the result of a comparison of the final grades of two groups of engineering students taught using two different instructional formats. As will be shown, the holistic teaching and learning environment helped those engineering students that are traditionally less successful.     

Why do organizations not learn from incidents? Bottlenecks,causes and conditions for a failure to effectively learn

If organizations would be able to learn more effectively from incidents that occurred in the past, future incidents and consequential injury or damage can be prevented. To improve learning from incidents, this study aimed to identify limiting factors, i.e. the causes of the failure to effectively learn. In seven organizations focus groups were held to discuss factors that according to employees contributed to the failure to learn. By use of a model of the learning from incidents process, the steps, where difficulties for learning arose, became visible, and the causes for these difficulties could be studied.     

Choosing creativity: the role of individual risk and ambiguity aversion on creative concept selection in engineering design

While creativity is often seen as an indispensable quality of engineering design, individuals often select conventional or previously successful options during the concept selection process due to the inherent risk associated with creative concepts and their inadvertent bias against creativity. However, little is actually known about what factors attribute to the promotion or filtering of these creative concepts during concept selection. To address this knowledge gap, an exploratory study was conducted with 38 undergraduate engineering students. This study was aimed at investigating the impact of individual risk aversion, ambiguity aversion, and student educational level on the selection and filtering of creative ideas during the concept selection process. The results from this study indicate that individuals’ ability to generate creative ideas is not significantly related to their preference for creative ideas during concept selection, but individual risk aversion and ambiguity aversion are significantly related to both creative concept selection and creative idea generation. Our results also revealed that first- and third-year students’ creative ability is affected differently by varying levels of tolerance for ambiguity. These results highlight the need for a more directed focus on creativity in engineering education in both concept creation and concept selection. These results also add to our understanding of creativity during concept selection and provide guidelines for enhancing the design process.     

Risk–incentives trade-off and outside options

Despite the prominence of the risk–incentives trade-off in the agency theory, empirical evidence provides at best a mixed result. In this paper, we argue that these mixed results may be attributed to the heterogeneity of risk aversion magnitude among the agents and the outside options available on the market. We show that when an increase in the uncertainty amplifies the riskiness of principal’s internal project more than the agent’s outside option, a positive risk–incentives relationship can be predicted when the internal project is sufficiently risky. On the other hand, when an uncertainty increase amplifies the riskiness of outside option more than the internal project, this positive risk–incentives relationship occurs if the internal project is of limited risk. These results hold irrespective of whether the agent’s risk aversion magnitude is publicly observable or privately known. Our analysis leads to a more fine-tuned empirical validation of agency theory and some testable hypotheses on the interactions between internal and external uncertainties.     

Pricing and promised delivery lead time decisions with a risk-averse agent

This paper studies how a risk-averse firm determines promised delivery lead time and price in a circumstance where the market demand depends not only on the selling price but also on the promised delivery lead time. Two scenarios, one in which the price and promised delivery lead time are independent and the other in which the price is promised delivery lead time sensitive, are analysed. Under conditional value-at-risk criterion, we examine how the risk aversion of decision-makers influences the optimal selection of price, promised delivery lead time and the overall utility of a risk-averse decision-maker. Moreover, by conducting comprehensive sensitivity analyses, we investigate the interaction effects of the decision-maker’s risk aversion and different parameters, including the price and lead time elasticity indexes, the inventory costs, as well as the capacity parameters on the optimal decisions and utilities, so as to offer managerial implications for practitioners under time-based competition environment.     

二阶随机占优约束下考虑订购能力的多产品报童问题

在不确定环境下的报童问题中，利用风险度量去对抗未来的不确定性，是帮助决策者规避风险的重要方法．二阶随机占优（Second order Stochastic Dominance, SSD）是一种稳健的风险度量．本文首先提出一种带有 SSD 约束及订购能力约束的风险厌恶多产品报童模型（SSD 模型）．其次，采用样本均值逼近（Sample Average Approximation, SAA）方法近似该问题，并对 SAA 问题进行收敛性分析．最后，在数值实验部分，用切平面法求解 SAA 问题，并同时与基于风险中性（无风险约束）和风险厌恶（以方差为风险约束）假设下的参照模型进行比较．数值结果表明相对于参照模型，在样本外预测下 SSD 模型可以更好地规避风险，得到更高的收益．     

Opposite valence social information provided by bio-robotic demonstrators shapes selection processes in the green bottle fly

Social learning represents a high-level complex process to acquire information about the environment, which is increasingly reported in invertebrates. The animal–robot interaction paradigm turned out to be an encouraging strategy to unveil social learning in vertebrates, but it has not been fully exploited in invertebrates. In this study, Lucilia sericata adults were induced to observe bio-robotic conspecific and predator demonstrators to reproduce different flower foraging choices. Can a fly manage two flows of social information with opposite valence? Herein, we attempt a reply. The selection process of L. sericata was affected by social information provided through different bio-robotic demonstrators, by avoiding coloured discs previously visited by a bio-robotic predator and preferring coloured discs previously visited by a bio-robotic conspecific. When both bio-robotic demonstrators visited the same disc, the latency duration increased and the flies significantly tended to avoid this disc. This indicates the complex risk–benefit evaluation process carried out by L. sericata during the acquisition of such social information. Overall, this article provides a unique perspective on the behavioural ecology of social learning in non-social insects; it also highlights the high potential of the animal–robot interaction approach for unveiling the full spectrum of invertebrates'' abilities in using social information.     

Strategies for sequential design of experiments and augmentation

The benefits of sequential design of experiments have long been described for both model-based and space-filling designs. However, in our experience, too few practitioners take advantage of the opportunity afforded by this approach to maximize the learning from their experimentation. By obtaining data sequentially, it is possible to learn from the early stages to inform subsequent data collection, minimize wasted resources, and provide answers for a series of objectives for the overall experiment. This paper provides methods and algorithms to create augmented distance-based space-filling designs, using both uniform and non-uniform space-filling strategies, that can be constructed at each stage based on information learned in earlier stages. We illustrate the methods with several examples that involve different initial data, types of space-filing designs and experimental goals.