Chapter 8. A Constraint-Based Approach to Learning and Reasoning

Neural-symbolic models bridge the gap between sub-symbolic and symbolic approaches, both of which have significant limitations. Sub-symbolic approaches, like neural networks, require a large amount of labeled data to be successful, whereas symbolic approaches, like logic reasoners, require a small amount of prior domain knowledge but do not easily scale to large collections of data. This chapter presents a general approach to integrate learning and reasoning that is based on the translation of the available prior knowledge into an undirected graphical model. Potentials on the graphical model are designed to accommodate dependencies among random variables by means of a set of trainable functions, like those computed by neural networks. The resulting neural-symbolic framework can effectively leverage the training data, when available, while exploiting high-level logic reasoning in a certain domain of discourse. Although exact inference is intractable within this model, different tractable models can be derived by making different assumptions. In particular, three models are presented in this chapter: Semantic-Based Regularization, Deep Logic Models and Relational Neural Machines. Semantic-Based Regularization is a scalable neural-symbolic model, that does not adapt the parameters of the reasoner, under the assumption that the provided prior knowledge is correct and must be exactly satisfied. Deep Logic Models preserve the scalability of Semantic-Based Regularization, while providing a flexible exploitation of logic knowledge by co-training the parameters of the reasoner during the learning procedure. Finally, Relational Neural Machines provide the fundamental advantages of perfectly replicating the effectiveness of training from supervised data of standard deep architectures, and of preserving the same generality and expressive power of Markov Logic Networks, when considering pure reasoning on symbolic data. The bonding between learning and reasoning is very general as any (deep) learner can be adopted, and any output structure expressed via First-Order Logic can be integrated. However, exact inference within a Relational Neural Machine is still intractable, and different factorizations are discussed to increase the scalability of the approach.

Exploring the implementation and underlying mechanisms of centralized referral systems to access specialized health services in Quebec

Background In 2016, Quebec, a Canadian province, implemented a program to improve access to specialized health services (Accès priorisé aux services spécialisés (APSS)), which includes single regional access points for processing requests to such services via primary care (Centre de répartition des demandes de services (CRDS)). Family physicians fill out and submit requests for initial consultations with specialists using a standardized form with predefined prioritization levels according to listed reasons for consultations, which is then sent to the centralized referral system (the CRDS) where consultations with specialists are assigned. We 1) described the APSS-CRDS program in three Quebec regions using logic models; 2) compared similarities and differences in the components and processes of the APSS-CRDS models; and 3) explored contextual factors influencing the models’ similarities and differences. Methods We relied on a qualitative study to develop logic models of the implemented APSS-CRDS program in three regions. Semi-structured interviews with health administrators (n = 9) were conducted. The interviews were analysed using a framework analysis approach according to the APSS-CRDS’s components included in the initially designed program, Mitchell and Lewis (2003)’s logic model framework, and Chaudoir and colleagues (2013)’s framework on contextual factors’ influence on an innovation’s implementation. Results Findings show the APSS-CRDS program’s regional variability in the implementation of its components, including its structure (centralized/decentralized), human resources involved in implementation and operation, processes to obtain specialists’ availability and assess/relay requests, as well as monitoring methods. Variability may be explained by contextual factors’ influence, like ministerial and medical associations’ involvement, collaborations, the context’s implementation readiness, physician practice characteristics, and the program’s adaptability. Interpretation Findings are useful to inform decision-makers on the design of programs like the APSS-CRDS, which aim to improve access to specialists, the essential components for the design of these types of interventions, and how contextual factors may influence program implementation. Variability in program design is important to consider as it may influence anticipated effects, a next step for the research team. Results may also inform stakeholders should they wish to implement similar programs to increase access to specialized health services via primary care.

Prediction of Unconfined Compressive Strength of Microfine Cement Injected Sands Using Fuzzy Logic Method

In this study, unconfined compressive strength values of sand soil injected with microfine cement were predicted using fuzzy logic method. Mamdani and Sugeno methods were applied in the fuzzy logic models. In addition, a regression analysis was carried out in order to compare these two methods. In the models, water/cement ratio and injection pressure were the input variables, and unconfined compressive strength was the output variable. The dataset includes 427 samples, which were experimentally injected with microfine cement. Predictions for unconfined compressive strength were obtained by creating membership functions and rule base for each input (predictive) parameter in fuzzy logic models. The coefficient of determination (R2) and Mean Square Error (MSE) were used as criteria for evaluating the performance of the developed models. The results suggested that the three applied models (i.e. Mamdani, Sugeno and regression) provided statistically significant results, and these methods could be used in the future prediction-based studies. The results showed that Sugeno model provided the best performance for predicting unconfined compressive strength. It was followed by Mamdani and Regression models, respectively. This study has suggested that the fuzzy logic method can be an alternative to the regression method which traditionally has been used in prediction process.

Developmental Evaluation for Extension Programs

Logic models have garnered acclaim for their usefulness and disdain for the time required to create good ones. We argue that the orderly, analytical nature of logic models is opposed to many Extension programs, and we explain developmental evaluation, an approach that highlights ongoing development, adaptations, and rapid response. We use our recently completed evaluation of the 4-H Science: Building a 4-H Career Pathway Initiative to demonstrate developmental evaluation’s key principles. Recommendations for Extension include the need to embrace developmental evaluation for program planning and evaluation and for Extension evaluators to conduct case studies using developmental evaluation and other approaches.

Five ways to get a grip on the drawbacks of logic models in program evaluation

Logic models are perhaps the most widely used tools in program evaluation work. They provide reasonably straightforward, visual illustrations of plausible links between program activities and outcomes. Consequently, they are employed frequently in stakeholder engagement, communication, and evaluation project planning. However, their relative simplicity comes with multiple drawbacks that can compromise the integrity of evaluation studies. In this Black Ice article, we outline key considerations and provide practical strategies that can help those engaged in evaluation work to identify and mitigate the limitations of logic models.

Citizen science impact pathways for a positive contribution to public participation in science

Positioning citizen science within the broader historical public engagement framework demonstrates how it has the potential to effectively tackle research and innovation issues. Citizen science approaches have their own challenges, which need to be considered in order to achieve this aim and contribute to wider and deeper public engagement. However, programme evaluations, which discuss lessons learned in engaging the public and other stakeholders with science are rare. To address this gap, we present the H2020-funded DITOs project and discuss the use of logic models in citizen science. We share the project’s assumptions, design considerations for deeper engagement and its impact pathways demonstrating how logic models can be utilised in citizen science to monitor programme effectiveness and for their successful implementation. We hope that this work will inspire citizen science practitioners to use similar tools and by doing so, share their experiences and potential barriers. This knowledge is essential for improving the way citizen science is currently practiced and its impacts to both science and society.

Marine Energy Environmental Permitting and Compliance Costs

Costs to permit Marine Energy projects are poorly understood. In this paper we examine environmental compliance and permitting costs for 19 projects in the U.S., covering the last 2 decades. Guided discussions were conducted with developers over a 3-year period to obtain historical and ongoing project cost data relative to environmental studies (e.g., baseline or pre-project site characterization as well as post-installation effects monitoring), stakeholder outreach, and mitigation, as well as qualitative experience of the permitting process. Data are organized in categories of technology type, permitted capacity, pre- and post-installation, geographic location, and funding types. We also compare our findings with earlier logic models created for the Department of Energy (i.e., Reference Models). Environmental studies most commonly performed were for Fish and Fisheries, Noise, Marine Habitat/Benthic Studies and Marine Mammals. Studies for tidal projects were more expensive than those performed for wave projects and the range of reported project costs tended to be wider than ranges predicted by logic models. For eight projects reporting full project costs, from project start to FERC or USACE permit, the average amount for environmental permitting compliance was 14.6%.