Umeå University's logo

The use of AI in social sciences is on the rise and gaining significant momentum. The book aims to demystify AI and its applications in social sciences by offering a straightforward description of its key concepts and tools, all in an easy-to-understand language. The book delves into the fundamental elements of AI systems, shedding light on their capabilities as well as their inherent constraints. Moreover, the author explores AI tools that find applications in the realm of social sciences, including decision support systems rooted in machine learning, as well as knowledge-based. In addition, the book addresses the pertinent issues that, while not purely AI-related, are crucial in AI applications within the social sciences, most notably security and privacy considerations. The content should be accessible to an audience ranging from undergraduate students to researchers and practitioners in the fields of AI and social sciences, assuming they possess a basic understanding of mathematics and a keen interest in the subject. This book is a bridge between the worlds of AI and social sciences, aiming to provide a fundamental understanding for those eager to explore the intersection of these two domains.

Fuzzy measures, also called capacities, non-additive measures, and monotonic games, are an interesting mathematical object and have been used in a large number of real-world applications. They are useful in problems where we need to aggregate information. In this case, we often use them in combination with fuzzy integrals. Sugeno and Choquet integrals are classical examples of fuzzy integrals.

Fuzzy measures are set functions, and as such, when the reference set is finite, they are defined by 2 n values where n represents the cardinality of the reference set. This makes difficult both to define them and to interpret them.

In this work, we propose to approximate fuzzy measures in terms of distorted probabilities as a way to better understand the structure and properties of the measure. This approach can be used when fuzzy measures are learned from examples. We show, however, that, in general, the approximation will not represent all the properties of an arbitrary fuzzy measure.

Microaggregation is a powerful technique for safeguarding data, enabling us to strike a balance between the risk of disclosing sensitive information and the loss of valuable information. It is a crucial tool for data sharing that provides k-anonymity. With the growing prevalence of multi-view data, there is an increasing interest in protecting such data using appropriate techniques. To the best of our knowledge, this paper introduces the first approach specifically designed for multi-view data protection. We present a novel approach to microaggregation by introducing multi-view fuzzy c-means, which allows us to consider linear constraints on the variables in each view that describe the data. Our method ensures that the resulting clusters adhere to these constraints, even when the data being masked fails to satisfy them. This approach not only enhances data privacy by maintaining k-anonymity in multi-view contexts but also preserves the structural integrity of the data across different views. Our contributions include the development of a multi-view clustering framework with built-in privacy safeguards and the introduction of linear constraints to ensure consistency across multiple views. This innovative approach provides a robust solution for the privacy-preserving analysis and sharing of multi-view data.

Machine Unlearning allows participants to remove their data from a trained machine learning model in order to preserve their privacy, and security. However, the machine unlearning literature for generative models is rather limited. The literature for image-to-image generative model (I2I model) considers minimizing the loss between Gaussian noise and the output of I2I model for forget samples as machine unlearning. However, we argue that the machine learning model performs fairly well on unseen data i.e., a retrained model will be able to learn generalized representations in the data and hence will not generate an output which is Gaussian noise instead. In this paper, we consider that the model after unlearning should treat forget samples as out-of-distribution (OOD) data, i.e., the unlearned model should no longer recognize or encode the specific patterns found in the forget samples. To achieve this, we propose a framework which decouples the model parameters with gradient ascent. Our framework ensures that forget samples are OOD for unlearned model with theoretical guarantee. We also provide (ϵ, δ)-unlearning guarantee for model updates with gradient ascent. The unlearned model is further fine-tuned on the remaining samples to maintain its performance. We also propose a data poisoning attack model as an auditing mechanism in order to make sure that the unlearned model has effectively removed the influence of forget samples. Furthermore, we demonstrate that even under sample unlearning, our approach prevents backdoor regeneration, validating its effectiveness. Extensive empirical evaluation on two large-scale datasets, ImageNet-1K and Places365 highlights the superiority of our approach. To show comparable performance with a retrained model, we also show the comparison of a simple AutoEncoder on various baselines on CIFAR-10 dataset.

We present functions that quantify the contribution of a set of arguments in quantitative bipolar argumentation graphs to (the final strength of) an argument of interest—a so-called topic. Our set contribution functions are generalizations of existing functions that quantify the contribution of a single contributing argument to a topic. Accordingly, we generalize existing contribution function principles for set contribution functions and provide a corresponding principle-based analysis. We introduce new principles specific to set-based functions that focus on properties pertaining to the interaction of arguments within a set. Finally, we sketch how the principles play out across different set contribution functions given a recommendation system application scenario.

Entropy is a fundamental concept in information theory but also in some AI algorithms, used for comparison of two distribution func-tions or measures that describe one specific event. When focusing on the differential (continuous) version of the entropy within the fuzzy measure framework, it is necessary to generalise all the essential concepts from the additive case to the fuzzy setup. This involves shifting from prob-ability to fuzzy measures, from Lebesgue to Choquet integral and from Radon-Nikodym to Choquet-Radon-Nikodym derivatives. In the paper, two formulas for defining fuzzy entropy are proposed with the use of extended versions of the Choquet integral. Their basic properties are examined and compared with the additive case, and a relation with the fuzzy Kullback-Leibler divergence is derived. Using a novel insight into derivatives known as the resulting measure approach, some computa-tions are presented to determine the final entropy value for two given measures.

Ensuring Real-Time Bidding compliance with GDPR and the ePrivacy Directive remains challenging. Existing solutions, like the IAB’s Transparency and Consent Framework and OpenRTB, lack transparency and fail to secure legally valid user consent. We propose a blockchain-based framework that decentralizes consent management, giving users direct control. The system automates consent handling, provides immutable compliance proof, and includes a smartphone app for users to manage consent per website. Additionally, we introduce a specificity value for IAB Audience Taxonomy elements, helping users assess the privacy impact of sharing data. By enhancing autonomy, transparency, and accountability, our approach strengthens trust in programmatic advertising while maintaining GDPR compliance without disrupting the industry.

Interactions among different elements of complex networks are organized in a structured manner. The collective behavior of the elements of these networks is organized according to community structure. Several methods have been defined to automatically detect these substructures in the field known as community discovery. Most of the methods have been applied to static or aggregated data. Recently the identification of evolving communities has gained more attention. Studying the relations among individuals yields insights on how communities form and evolve, but there are some limits that should be enforced to respect individuals’ privacy while sharing and collecting their data. Privacy-protection techniques have been commonly applied to static data, while there are few methods that work on dynamic data. Recently, there have been some approaches to protect dynamic graphs with local edge-differential privacy that have been tested for community discovery applications. However, the evolution of the communities over time has not been evaluated on the privacy-protected data. We test the utility considering community discovery and evolution in time-varying networks for such localedge-ϵ-differential privacy methods. We show empirically how these algorithms can provide privacy while preserving the community lifecycles, for their privacy-aware study.

Deep neural networks (DNNs) are one of the most widely used machine learning algorithm. DNNs requires the training data to be available beforehand with true labels. This is not feasible for many real-world problems where data arrives in the streaming form and acquisition of true labels are scarce and expensive. In the literature, not much focus has been given to the privacy prospect of the streaming data, where data may change its distribution frequently. These concept drifts must be detected privately in order to avoid any disclosure risk from DNNs. Existing privacy models use concept drift detection schemes such ADWIN, KSWIN to detect the drifts. In this paper, we focus on the notion of integrally private DNNs to detect concept drifts. Integrally private DNNs are the models which recur frequently from different datasets. Based on this, we introduce an ensemble methodology which we call 'Integrally Private Drift Detection' (IPDD) method to detect concept drift from private models. Our IPDD method does not require labels to detect drift but assumes true labels are available once the drift has been detected. We have experimented with binary and multi-class synthetic and real-world data. Our experimental results show that our methodology can privately detect concept drift, has comparable utility (even better in some cases) with ADWIN and outperforms utility from different levels of differentially private models.

Machine and statistical learning mainly consists of first deciding on a parametric model and then fitting its corresponding parameters. For this model fitting, it is usual to consider a loss function. There are in the literature several types of models based on non-additive measures. The most common examples include models based on fuzzy integrals (as e.g., Choquet and Sugeno integrals). In this case, given some data, a measure is identified and the model is built. That is, we build a data-driven model based on a non-additive measure (or on several nonadditive measures). Then, once the measure is identified, we are often interested in their analysis and visualization to understand its properties. In this paper we give an overview of measure identification and measure analysis.