Search

The purpose of this paper is to propose a novel hybrid framework for evaluating and benchmarking trustworthy artificial intelligence (AI) applications in healthcare by using multi-criteria decision-making (MCDM) techniques under a new fuzzy environment. To develop such a framework, a new decision matrix has been built, and then integrated with q-ROF2TL-FWZIC (q‐Rung Orthopair Fuzzy 2‐Tuple Linguistic Fuzzy-Weighted Zero-Inconsistency) and q-ROF2TL-CODAS (q‐Rung Orthopair Fuzzy 2‐Tuple Linguistic Combinative Distance-Based Assessment). In this integration, q-ROF2TL-FWZIC is utilized for assigning the weights of evaluation attributes of trustworthy AI, while q-ROF2TL-CODAS is employed for benchmarking trustworthy AI applications. Findings show that the q-ROF2TL-FWZIC method effectively weights the evaluation attributes. The transparency attribute receives the highest importance weight (0.173566825), whereas the human agency and oversight criterion has the lowest weight (0.105741901). The remaining attributes are distributed in between. Moreover, alternative_4 receives the highest rank order (score of 7.370410417), while alternative_13 receives the lowest rank order (score of −4.759794397). To evaluate the validity of the proposed framework, systematic ranking and sensitivity analysis assessments were employed. [ABSTRACT FROM AUTHOR]

Intelligent computing is an independent algorithm system that imitates the laws and principles of nature's biological chain to design and solve practical problems in modern society, it can help people better handle some big data problems through artificial intelligence technology and different cognitive characteristics. Especially, when dealing with group decision problems, it is often considered as human behavior based on cognitive computing. The q-rung orthopair fuzzy set (q-ROFS) is not only a generalization of Pythagorean fuzzy set (PFS), but also can deal with the decision problem of multi-source information in a wider area. In this paper, we first give the detailed explanations of the relationship between the q-ROFS and q-rung orthopair fuzzy number (q-ROFN)by an example, and itis pointed out that there are some defects in the existing two ranking methods for the Pythagorean fuzzy numbers (PFNs) and intuitionistic fuzzy numbers (IFNs) through counterexamples, and then all q-ROFNs are mapped into the unit triangle of the first quadrant through q-compression mapping, and the concept of distance factor for q-ROFNs is proposed by Euclidean distance between each q-ROFN and the maximum element (1,0).Secondly, a novel inverse score function for q-ROFNs are put forward by the distance factor and hesitancy degree, and the monotonicity of the inverse score function are discussed. Finally, a new ranking criterion of q-ROFNs is given by the new score formula, and then the rationality of the ranking criterion is proved, and through the method of solving the conditional extreme value of multivariate function (Lagrange multiplier method), there is indeed an extreme value point (corresponding to the maximum q-ROFN) on υ = μ. Besides, a q-rung orthopair fuzzy decision-making method for multi-source information is given by the new ranking criterion and aggregation operations, and the comparison of examples shows that the inverse score function and decision-making method do have many advantages in the q-ROFN environment. [ABSTRACT FROM AUTHOR]

Due to the increasing use of Machine Learning models in high stakes decision making settings, it has become increasingly important to have tools to understand how models arrive at decisions. Assuming an already trained Supervised Classification model, post-hoc explanations can be obtained via so-called counterfactual analysis: a counterfactual explanation of an instance indicates how this instance should be minimally modified so that the perturbed instance is classified in the desired class by the given Machine Learning classification model. Most of the Counterfactual Analysis literature focuses on the single-instance single-counterfactual setting, in which the analysis is done for one single instance to provide one single counterfactual explanation. Taking a stakeholder's perspective, in this paper we introduce the so-called collective counterfactual explanations. By means of novel Mathematical Optimization models, we provide a counterfactual explanation for each instance in a group of interest, so that the total cost of the perturbations is minimized under some linking constraints. Making the process of constructing counterfactuals collective instead of individual enables us to detect the features that are critical to the entire dataset to have the individuals classified in the desired class. Our methodology allows for some instances to be treated individually, as in the single-instance single-counterfactual case, performing the collective counterfactual analysis for a fraction of records of the group of interest. This way, outliers are identified and handled appropriately. Under some assumptions on the classifier and the space in which counterfactuals are sought, finding collective counterfactual explanations is reduced to solving a convex quadratic linearly constrained mixed integer optimization problem, which, for datasets of moderate size, can be solved to optimality using existing solvers. The performance of our approach is illustrated on real-world datasets, demonstrating its usefulness. • A new model for Explainable Artificial Intelligence is proposed. • Counterfactual Explanations for a full group of individuals are sought. • Sparsity both at individual and group level is achieved. [ABSTRACT FROM AUTHOR]

Maritime Autonomous Surface Ships (MASS) have obtained wide attention in recent years, which is relied on their attractive socio-economic benefits and potential to improve safety. One of the main characteristics of MASS is that it can use perception information to replace Officers on Watch (OOW) to make different navigation decisions through expert and intelligent systems, so as to make navigation more intelligent, reliable and safer, which is mainly based on the Maritime Autonomous Navigation Decision-making System (MANDS). However, the state-of-the-art study lacks a systematic description and research on MANDS, and most of them mainly focus on solving a single navigation decision-making task, which is not aligned with the meaning of autonomous navigation. Therefore, this study develops a novel MANDS and installs it on a 300TEU intelligent container ship to successfully carry out real ship trials of autonomous navigation. In this paper, the definition of autonomous navigation is first proposed, which mainly contains four different navigation tasks, namely "Keep route", "Route Optimization", "Collision Avoidance", "Return Back". A well-designed architecture for MANDS is presented which includes corresponding decision-making models and communication interfaces with external systems. Furthermore, in order to realize the switching and deployment between different navigation tasks in the process of autonomous navigation, a novel hierarchical decision-making mechanism is proposed and analyzed, and finally the MANDS is integrated with perception system and ship control system respectively. Before the real ship trial, this system is tested and verified in a simulation-based environment which is composed of the electronic chart platform and ship motion simulator, we also design a scenario library for the testing work. At last, real ship trials of autonomous navigation for MANDS are carried out, which can convincingly prove the performance of MANDS. This study proposes a relative complete solution for autonomous navigation, including system modelling, integration and real ship trial. These findings offer new insight into the research of ship autonomous navigation and promote the improvement of ship intelligence and informatization level. [ABSTRACT FROM AUTHOR]

• Uncertainty in decision-making systems by categorical data. • Preprocessing incomplete and ambiguous categorical attributes. • Uncertainty due to lack of pre-modelling explainability. • Subjective expert preference over a piece of evidence in categorical attribute. The uncertainty attributed by discrepant data in AI-enabled decisions is a critical challenge in highly regulated domains such as health care and finance. Ambiguity and incompleteness due to missing values in output and input attributes, respectively, is ubiquitous in these domains. It could have an adverse impact on a certain unrepresented set of people in the training data without a developer's intention to discriminate. The inherently non-numerical nature of categorical attributes than numerical attributes and the presence of incomplete and ambiguous categorical attributes in a dataset increases the uncertainty in decision-making. This paper addresses the challenges in handling categorical attributes as it is not addressed comprehensively in previous research. Three sources of uncertainties in categorical attributes are recognised in this research. The informational uncertainty, unforeseeable uncertainty in the decision task environment, and the uncertainty due to lack of pre-modelling explainability in categorical attributes are addressed in the proposed methodology on maximum likelihood evidential reasoning (MAKER). It can transform and impute incomplete and ambiguous categorical attributes into interpretable numerical features. It utilises a notion of weight and reliability to include subjective expert preference over a piece of evidence and the quality of evidence in a categorical attribute, respectively. The MAKER framework strives to integrate the recognised uncertainties in the transformed input data that allow a model to perceive data limitations during the training regime and acknowledge doubtful predictions by supporting trustworthy pre-modelling and post modelling explainability. The ability to handle uncertainty and its impact on explainability is demonstrated on a real-world healthcare and finance data for different missing data scenarios in three types of AI algorithms: deep-learning, tree-based, and rule-based model. [ABSTRACT FROM AUTHOR]