| 1. |
- Abril, Daniel, et al.
(författare)
-
Spherical Microaggregation : Anonymizing Sparse Vector Spaces
- 2015
-
Ingår i: Computers & security (Print). - : Elsevier. - 0167-4048 .- 1872-6208. ; 49, s. 28-44
-
Tidskriftsartikel (refereegranskat)abstract
- Unstructured texts are a very popular data type and still widely unexplored in the privacy preserving data mining field. We consider the problem of providing public information about a set of confidential documents. To that end we have developed a method to protect a Vector Space Model (VSM), to make it public even if the documents it represents are private. This method is inspired by microaggregation, a popular protection method from statistical disclosure control, and adapted to work with sparse and high dimensional data sets.
|
|
| 2. |
- Abril, Daniel, et al.
(författare)
-
Supervised Learning Using a Symmetric Bilinear Form for Record Linkage
- 2015
-
Ingår i: Information Fusion. - : Elsevier. - 1566-2535 .- 1872-6305. ; 26, s. 144-153
-
Tidskriftsartikel (refereegranskat)abstract
- Record Linkage is used to link records of two different files corresponding to the same individuals. These algorithms are used for database integration. In data privacy, these algorithms are used to evaluate the disclosure risk of a protected data set by linking records that belong to the same individual. The degree of success when linking the original (unprotected data) with the protected data gives an estimation of the disclosure risk.In this paper we propose a new parameterized aggregation operator and a supervised learning method for disclosure risk assessment. The parameterized operator is a symmetric bilinear form and the supervised learning method is formalized as an optimization problem. The target of the optimization problem is to find the values of the aggregation parameters that maximize the number of re-identification (or correct links). We evaluate and compare our proposal with other non-parametrized variations of record linkage, such as those using the Mahalanobis distance and the Euclidean distance (one of the most used approaches for this purpose). Additionally, we also compare it with other previously presented parameterized aggregation operators for record linkage such as the weighted mean and the Choquet integral. From these comparisons we show how the proposed aggregation operator is able to overcome or at least achieve similar results than the other parameterized operators. We also study which are the necessary optimization problem conditions to consider the described aggregation functions as metric functions.
|
|
| 3. |
- Adewole, Kayode S., et al.
(författare)
-
DFTMicroagg: a dual-level anonymization algorithm for smart grid data
- 2022
-
Ingår i: International Journal of Information Security. - : Springer. - 1615-5262 .- 1615-5270. ; 21, s. 1299-1321
-
Tidskriftsartikel (refereegranskat)abstract
- The introduction of advanced metering infrastructure (AMI) smart meters has given rise to fine-grained electricity usage data at different levels of time granularity. AMI collects high-frequency daily energy consumption data that enables utility companies and data aggregators to perform a rich set of grid operations such as demand response, grid monitoring, load forecasting and many more. However, the privacy concerns associated with daily energy consumption data has been raised. Existing studies on data anonymization for smart grid data focused on the direct application of perturbation algorithms, such as microaggregation, to protect the privacy of consumers. In this paper, we empirically show that reliance on microaggregation alone is not sufficient to protect smart grid data. Therefore, we propose DFTMicroagg algorithm that provides a dual level of perturbation to improve privacy. The algorithm leverages the benefits of discrete Fourier transform (DFT) and microaggregation to provide additional layer of protection. We evaluated our algorithm on two publicly available smart grid datasets with millions of smart meters readings. Experimental results based on clustering analysis using k-Means, classification via k-nearest neighbor (kNN) algorithm and mean hourly energy consumption forecast using Seasonal Auto-Regressive Integrated Moving Average with eXogenous (SARIMAX) factors model further proved the applicability of the proposed method. Our approach provides utility companies with more flexibility to control the level of protection for their published energy data.
|
|
| 4. |
- Adewole, Kayode Sakariyah, et al.
(författare)
-
Energy disaggregation risk resilience through microaggregation and discrete Fourier transform
- 2024
-
Ingår i: Information Sciences. - : Elsevier. - 0020-0255 .- 1872-6291. ; 662
-
Tidskriftsartikel (refereegranskat)abstract
- Progress in the field of Non-Intrusive Load Monitoring (NILM) has been attributed to the rise in the application of artificial intelligence. Nevertheless, the ability of energy disaggregation algorithms to disaggregate different appliance signatures from aggregated smart grid data poses some privacy issues. This paper introduces a new notion of disclosure risk termed energy disaggregation risk. The performance of Sequence-to-Sequence (Seq2Seq) NILM deep learning algorithm along with three activation extraction methods are studied using two publicly available datasets. To understand the extent of disclosure, we study three inference attacks on aggregated data. The results show that Variance Sensitive Thresholding (VST) event detection method outperformed the other two methods in revealing households' lifestyles based on the signature of the appliances. To reduce energy disaggregation risk, we investigate the performance of two privacy-preserving mechanisms based on microaggregation and Discrete Fourier Transform (DFT). Empirically, for the first scenario of inference attack on UK-DALE, VST produces disaggregation risks of 99%, 100%, 89% and 99% for fridge, dish washer, microwave, and kettle respectively. For washing machine, Activation Time Extraction (ATE) method produces a disaggregation risk of 87%. We obtain similar results for other inference attack scenarios and the risk reduces using the two privacy-protection mechanisms.
|
|
| 5. |
- Adewole, Kayode Sakariyah, et al.
(författare)
-
Privacy issues in smart grid data : from energy disaggregation to disclosure risk
- 2022
-
Ingår i: Database and expert systems applications. - Cham : Springer. - 9783031124228 - 9783031124235 ; , s. 71-84
-
Konferensbidrag (refereegranskat)abstract
- The advancement in artificial intelligence (AI) techniques has given rise to the success rate recorded in the field of Non-Intrusive Load Monitoring (NILM). The development of robust AI and machine learning algorithms based on deep learning architecture has enabled accurate extraction of individual appliance load signature from aggregated energy data. However, the success rate of NILM algorithm in disaggregating individual appliance load signature in smart grid data violates the privacy of the individual household lifestyle. This paper investigates the performance of Sequence-to-Sequence (Seq2Seq) deep learning NILM algorithm in predicting the load signature of appliances. Furthermore, we define a new notion of disclosure risk to understand the risk associated with individual appliances in aggregated signals. Two publicly available energy disaggregation datasets have been considered. We simulate three inference attack scenarios to better ascertain the risk of publishing raw energy data. In addition, we investigate three activation extraction methods for appliance event detection. The results show that the disclosure risk associated with releasing smart grid data in their original form is on the high side. Therefore, future privacy protection mechanisms should devise efficient methods to reduce this risk.
|
|
| 6. |
- Adewole, Kayode Sakariyah, et al.
(författare)
-
Privacy protection of synthetic smart grid data simulated via generative adversarial networks
- 2023
-
Ingår i: Proceedings of the 20th international conference on security and cryptography, SECRYPT 2023. - : SciTePress. - 9789897586668 ; , s. 279-286
-
Konferensbidrag (refereegranskat)abstract
- The development in smart meter technology has made grid operations more efficient based on fine-grained electricity usage data generated at different levels of time granularity. Consequently, machine learning algorithms have benefited from these data to produce useful models for important grid operations. Although machine learning algorithms need historical data to improve predictive performance, these data are not readily available for public utilization due to privacy issues. The existing smart grid data simulation frameworks generate grid data with implicit privacy concerns since the data are simulated from a few real energy consumptions that are publicly available. This paper addresses two issues in smart grid. First, it assesses the level of privacy violation with the individual household appliances based on synthetic household aggregate loads consumption. Second, based on the findings, it proposes two privacy-preserving mechanisms to reduce this risk. Three inference attacks are simulated and the results obtained confirm the efficacy of the proposed privacy-preserving mechanisms.
|
|
| 7. |
- Alcantud, Jose Carlos R., et al.
(författare)
-
Decomposition theorems and extension principles for hesitant fuzzy sets
- 2018
-
Ingår i: Information Fusion. - : Elsevier. - 1566-2535 .- 1872-6305. ; 41, s. 48-56
-
Tidskriftsartikel (refereegranskat)abstract
- We prove a decomposition theorem for hesitant fuzzy sets, which states that every typical hesitant fuzzy set on a set can be represented by a well-structured family of fuzzy sets on that set. This decomposition is expressed by the novel concept of hesitant fuzzy set associated with a family of hesitant fuzzy sets, in terms of newly defined families of their cuts. Our result supposes the first representation theorem of hesitant fuzzy sets in the literature. Other related representation results are proven. We also define two novel extension principles that extend crisp functions to functions that map hesitant fuzzy sets into hesitant fuzzy sets.
|
|
| 8. |
- Aliahmadipour, Laya, et al.
(författare)
-
A definition for hesitant fuzzy partitions
- 2016
-
Ingår i: International Journal of Computational Intelligence Systems. - : Taylor & Francis Group. - 1875-6891 .- 1875-6883. ; 9:3, s. 497-505
-
Tidskriftsartikel (refereegranskat)abstract
- In this paper, we define hesitant fuzzy partitions (H-fuzzy partitions) to consider the results of standard fuzzy clustering family (e.g. fuzzy c-means and intuitionistic fuzzy c-means). We define a method to construct H-fuzzy partitions from a set of fuzzy clusters obtained from several executions of fuzzy clustering algorithms with various initialization of their parameters. Our purpose is to consider some local optimal solutions to find a global optimal solution also letting the user to consider various reliable membership values and cluster centers to evaluate her/his problem using different cluster validity indices.
|
|
| 9. |
- Aliahmadipour, L., et al.
(författare)
-
HFC : data clustering based on hesitant fuzzy decision making
- 2022
-
Ingår i: Iranian Journal of Fuzzy Systems. - : University of Sistan and Baluchestan. - 1735-0654 .- 2676-4334. ; 19:5, s. 167-181
-
Tidskriftsartikel (refereegranskat)abstract
- In a clustering task, choosing a proper clustering algorithm and obtaining qualified clusters are crucial issues. Sometimes, a clustering algorithm is chosen based on the data distribution, but data distributions are not known beforehand in real world problems. In this case, we hesitate which clustering algorithm to choose. In this paper, this hesitation is modeled by a hesitant fuzzy multi criteria decision making problem (HFMCDM) in which some clustering algorithms play the role of experts. Here, we consider fuzzy C-means (FCM) and agglomerative clustering algorithms as representative of two popular categories of clustering algorithms partitioning and hierarchical clustering methods, respectively.Then, we propose a new clustering procedure based on hesitant fuzzy decision making approaches (HFC) to decide which of the FCM family or hierarchical clustering algorithms is suitable for our data. This procedure ascertains a good clustering algorithm using neutrosophic FCM (NFCM) through a two phases process. The HFC procedure not only makes a true decision about applying partitioning clustering algorithms, but also improves the performance of FCM and evolutionary kernel intuitionistic fuzzy c-means clustering algorithm (EKIFCM) with construction hesitant fuzzy partition (HFP) conveniently. Experimental results show that the clustering procedure is applicable and practical. According to HFC procedure, it should be mentioned that it is possible to replace the other clustering algorithms that belong to any partitioning and hierarchical clustering methods. Also, we can consider other categories of clustering algorithms.
|
|
| 10. |
- Aliahmadipour, Laya, et al.
(författare)
-
On Hesitant Fuzzy Clustering and Clustering of Hesitant Fuzzy Data
- 2017
-
Ingår i: Fuzzy sets, rough sets, multisets and clustering. - Cham : Springer. - 9783319475561 - 9783319475578 ; , s. 157-168
-
Bokkapitel (refereegranskat)abstract
- Since the notion of hesitant fuzzy set was introduced, some clustering algorithms have been proposed to cluster hesitant fuzzy data. Beside of hesitation in data, there is some hesitation in the clustering (classification) of a crisp data set. This hesitation may be arise in the selection process of a suitable clustering (classification) algorithm and initial parametrization of a clustering (classification) algorithm. Hesitant fuzzy set theory is a suitable tool to deal with this kind of problems. In this study, we introduce two different points of view to apply hesitant fuzzy sets in the data mining tasks, specially in the clustering algorithms.
|
|
| 11. |
- Anjomshoae, Sule, 1985-
(författare)
-
Context-based explanations for machine learning predictions
- 2022
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- In recent years, growing concern regarding trust in algorithmic decision-making has drawn attention to more transparent and interpretable models. Laws and regulations are moving towards requiring this functionality from information systems to prevent unintended side effects. Such as the European Union's General Data Protection Regulations (GDPR) set out the right to be informed regarding machine-generated decisions. Individuals affected by these decisions can question, confront and challenge the inferences automatically produced by machine learning models. Consequently, such matters necessitate AI systems to be transparent and explainable for various practical applications.Furthermore, explanations help evaluate these systems' strengths and limitations, thereby fostering trustworthiness. As important as it is, existing studies mainly focus on creating mathematically interpretable models or explaining black-box algorithms with intrinsically interpretable surrogate models. In general, these explanations are intended for technical users to evaluate the correctness of a model and are often hard to interpret by general users. Given a critical need for methods that consider end-user requirements, this thesis focuses on generating intelligible explanations for predictions made by machine learning algorithms. As a starting point, we present the outcome of a systematic literature review of the existing research on generating and communicating explanations in goal-driven eXplainable AI (XAI), such as agents and robots. These are known for their ability to communicate their decisions in human understandable terms. Influenced by that, we discuss the design and evaluation of our proposed explanation methods for black-box algorithms in different machine learning applications, including image recognition, scene classification, and disease prediction.Taken together, the methods and tools presented in this thesis could be used to explain machine learning predictions or as a baseline to compare to other explanation techniques, enabling interpretation indicators for experts and non-technical users. The findings would also be of interest to domains using machine learning models for high-stake decision-making to investigate the practical utility of proposed explanation methods.
|
|
| 12. |
- Armengol, Eva, et al.
(författare)
-
Generalization-Based k-Anonymization
- 2015
-
Ingår i: Modeling Decisions for Artificial Intelligence. - Cham : Springer. - 9783319232393 - 9783319232409 ; , s. 207-218
-
Konferensbidrag (refereegranskat)abstract
- Microaggregation is an anonymization technique consistingon partitioning the data into clusters no smaller thankelements andthen replacing the whole cluster by its prototypical representant. Mostof microaggregation techniques work on numerical attributes. However,many data sets are described by heterogeneous types of data, i.e., nu-merical and categorical attributes. In this paper we propose a new mi-croaggregation method for achieving a compliantk-anonymous maskedfile for categorical microdata based on generalization. The goal is to builda generalized description satisfied by at leastkdomain objects and toreplace these domain objects by the description. The way to constructthat generalization is similar that the one used in growing decision trees.Records that cannot be generalized satisfactorily are discarded, thereforesome information is lost. In the experiments we performed we prove thatthe new approach gives good results.
|
|
| 13. |
- Armengol, Eva, et al.
(författare)
-
Partial Domain Theories for Privacy
- 2016
-
Ingår i: Modeling Decisions for Artificial Intelligence. - Cham : Springer. - 9783319456553 - 9783319456560 ; , s. 217-226
-
Konferensbidrag (refereegranskat)abstract
- Generalization and Suppression are two of the most used techniques to achieve k-anonymity. However, the generalization concept is also used in machine learning to obtain domain models useful for the classification task, and the suppression is the way to achieve such generalization. In this paper we want to address the anonymization of data preserving the classification task. What we propose is to use machine learning methods to obtain partial domain theories formed by partial descriptions of classes. Differently than in machine learning, we impose that such descriptions be as specific as possible, i.e., formed by the maximum number of attributes. This is achieved by suppressing some values of some records. In our method, we suppress only a particular value of an attribute in only a subset of records, that is, we use local suppression. This avoids one of the problems of global suppression that is the loss of more information than necessary.
|
|
| 14. |
- Bae, Juhee, et al.
(författare)
-
Complex Data Analysis
- 2019
-
Ingår i: Data science in Practice. - Cham : Springer. - 9783319975566 - 9783319975559 ; , s. 157-169
-
Bokkapitel (refereegranskat)abstract
- Data science applications often need to deal with data that does not fit into the standard entity-attribute-value model. In this chapter we discuss three of these other types of data. We discuss texts, images and graphs. The importance of social media is one of the reason for the interest on graphs as they are a way to represent social networks and, in general, any type of interaction between people. In this chapter we present examples of tools that can be used to extract information and, thus, analyze these three types of data. In particular, we discuss topic modeling using a hierarchical statistical model as a way to extract relevant topics from texts, image analysis using convolutional neural networks, and measures and visual methods to summarize information from graphs.
|
|
| 15. |
- Bae, Juhee, et al.
(författare)
-
On the Visualization of Discrete Non-additive Measures
- 2018
-
Ingår i: Aggregation Functions in Theory and in Practice AGOP 2017. - Cham : Springer. - 9783319593067 - 9783319593050 ; , s. 200-210
-
Konferensbidrag (refereegranskat)abstract
- Non-additive measures generalize additive measures, and have been utilized in several applications. They are used to represent different types of uncertainty and also to represent importance in data aggregation. As non-additive measures are set functions, the number of values to be considered grows exponentially. This makes difficult their definition but also their interpretation and understanding. In order to support understability, this paper explores the topic of visualizing discrete non-additive measures using node-link diagram representations.
|
|
| 16. |
- Bozorgpanah, Aso, et al.
(författare)
-
Explainable machine learning models with privacy
- 2024
-
Ingår i: Progress in Artificial Intelligence. - : Springer. - 2192-6352 .- 2192-6360. ; 13, s. 31-50
-
Tidskriftsartikel (refereegranskat)abstract
- The importance of explainable machine learning models is increasing because users want to understand the reasons behind decisions in data-driven models. Interpretability and explainability emerge from this need to design comprehensible systems. This paper focuses on privacy-preserving explainable machine learning. We study two data masking techniques: maximum distance to average vector (MDAV) and additive noise. The former is for achieving k-anonymity, and the second uses Laplacian noise to avoid record leakage and provide a level of differential privacy. We are interested in the process of developing data-driven models that, at the same time, make explainable decisions and are privacy-preserving. That is, we want to avoid the decision-making process leading to disclosure. To that end, we propose building models from anonymized data. More particularly, data that are k-anonymous or that have been anonymized add an appropriate level of noise to satisfy some differential privacy requirements. In this paper, we study how explainability has been affected by these data protection procedures. We use TreeSHAP as our technique for explainability. The experiments show that we can keep up to a certain degree both accuracy and explainability. So, our results show that some trade-off between privacy and explainability is possible for data protection using k-anonymity and noise addition.
|
|
| 17. |
- Bozorgpanah, Aso, et al.
(författare)
-
Privacy and explainability : the effects of data protection on shapley values
- 2022
-
Ingår i: Technologies. - : MDPI. - 2227-7080. ; 10:6
-
Tidskriftsartikel (refereegranskat)abstract
- There is an increasing need to provide explainability for machine learning models. There are different alternatives to provide explainability, for example, local and global methods. One of the approaches is based on Shapley values. Privacy is another critical requirement when dealing with sensitive data. Data-driven machine learning models may lead to disclosure. Data privacy provides several methods for ensuring privacy. In this paper, we study how methods for explainability based on Shapley values are affected by privacy methods. We show that some degree of protection still permits to maintain the information of Shapley values for the four machine learning models studied. Experiments seem to indicate that among the four models, Shapley values of linear models are the most affected ones.
|
|
| 18. |
- Casas-Roma, Jordi, et al.
(författare)
-
A survey of graph-modification techniques for privacy-preserving on networks
- 2017
-
Ingår i: Artificial Intelligence Review. - : Springer. - 0269-2821 .- 1573-7462. ; 47:3, s. 341-366
-
Tidskriftsartikel (refereegranskat)abstract
- Recently, a huge amount of social networks have been made publicly available. In parallel, several definitions and methods have been proposed to protect users’ privacy when publicly releasing these data. Some of them were picked out from relational dataset anonymization techniques, which are riper than network anonymization techniques. In this paper we summarize privacy-preserving techniques, focusing on graph-modification methods which alter graph’s structure and release the entire anonymous network. These methods allow researchers and third-parties to apply all graph-mining processes on anonymous data, from local to global knowledge extraction.
|
|
| 19. |
- Casas-Roma, Jordi, et al.
(författare)
-
k-Degree anonymity and edge selection : improving data utility in large networks
- 2017
-
Ingår i: Knowledge and Information Systems. - : Springer. - 0219-1377 .- 0219-3116. ; 50:2, s. 447-474
-
Tidskriftsartikel (refereegranskat)abstract
- The problem of anonymization in large networks and the utility of released data are considered in this paper. Although there are some anonymization methods for networks, most of them cannot be applied in large networks because of their complexity. In this paper, we devise a simple and efficient algorithm for k-degree anonymity in large networks. Our algorithm constructs a k-degree anonymous network by the minimum number of edge modifications. We compare our algorithm with other well-known k-degree anonymous algorithms and demonstrate that information loss in real networks is lowered. Moreover, we consider the edge relevance in order to improve the data utility on anonymized networks. By considering the neighbourhood centrality score of each edge, we preserve the most important edges of the network, reducing the information loss and increasing the data utility. An evaluation of clustering processes is performed on our algorithm, proving that edge neighbourhood centrality increases data utility. Lastly, we apply our algorithm to different large real datasets and demonstrate their efficiency and practical utility.
|
|
| 20. |
|
|
| 21. |
|
|
| 22. |
- Dujmović, Jozo, et al.
(författare)
-
Aggregation functions in decision engineering : ten necessary properties and parameter-directedness
- 2022
-
Ingår i: Intelligent and fuzzy techniques for emerging conditions and digital transformation. - Cham : Springer. - 9783030856250 - 9783030856267 ; , s. 173-181
-
Konferensbidrag (refereegranskat)abstract
- Applications, as, for example, decision support systems, need to combine information. The functions that permit to combine a set of numerical values into their single representative are called aggregation operators. They go from simple and well-known functions, as the arithmetic mean, to more sophisticated and complex, as fuzzy integrals. In the context of decision engineering and other applications, aggregators are used to develop methods and software tools that can be efficiently used in specific real life problem solving. In this context, when a new aggregator-based application is in development, we face the problem of deciding which aggregation functions and which parameters we are going to use. We review some results related to selecting engineering aggregation functions for real-world applications. We review ten necessary properties and discuss problems of parameter-directedness. Parameter-directedness is the request for explicit visibility and easy adjustability of all aggregator design and performance parameters. An example is the andness/orness-directedness.
|
|
| 23. |
- Dujmović, Jozo, et al.
(författare)
-
Logic aggregators and their implementations
- 2023
-
Ingår i: Modeling decisions for artificial intelligence. - : Springer Science+Business Media B.V.. - 9783031334979 - 9783031334986 ; , s. 3-42
-
Konferensbidrag (refereegranskat)abstract
- In this paper we present necessary properties of logic aggregators and compare their major implementations. If decision making includes the identification of a set of alternatives followed by the evaluation of alternatives and selection of the best alternative, then evaluation must be based on graded logic aggregation. The resulting analytic framework is a graded logic which is a seamless generalization of Boolean logic, based on analytic models of graded simultaneity (various forms of conjunction), graded substitutability (various forms of disjunction) and complementing (negation). These basic logic operations can be implemented in various ways, including means, t-norms/conorms, OWA, and fuzzy integrals. Such mathematical models must be applicable in all regions of the unit hypercube [ 0, 1 ] n. In order to be applicable in various areas of decision making, the logic aggregators must be consistent with observable patterns of human reasoning, supporting both formal logic and semantic aspects of human reasoning. That creates a comprehensive set of logic requirements that logic aggregators must satisfy. Various popular aggregators satisfy these requirements to the extent investigated in this paper. The results of our investigation clearly show the limits of applicability of the analyzed aggregators in the area of decision making.
|
|
| 24. |
- Eftekhari, Mahdi, et al.
(författare)
-
How fuzzy concepts contribute to machine learning
- 2022
-
Bok (refereegranskat)abstract
- This book introduces some contemporary approaches on the application of fuzzy and hesitant fuzzy sets in machine learning tasks such as classification, clustering and dimension reduction. Many situations arise in machine learning algorithms in which applying methods for uncertainty modeling and multi-criteria decision making can lead to a better understanding of algorithms behavior as well as achieving good performances. Specifically, the present book is a collection of novel viewpoints on how fuzzy and hesitant fuzzy concepts can be applied to data uncertainty modeling as well as being used to solve multi-criteria decision making challenges raised in machine learning problems. Using the multi-criteria decision making framework, the book shows how different algorithms, rather than human experts, are employed to determine membership degrees. The book is expected to bring closer the communities of pure mathematicians of fuzzy sets and data scientists.
|
|
| 25. |
- Fukushima, Takuya, et al.
(författare)
-
Optimal value estimation of intentional-value-substitution for learning regression models
- 2021
-
Ingår i: Journal of Advanced Computational Intelligence and Intelligent Informatics. - : Fuji Technology Press. - 1343-0130 .- 1883-8014. ; 25:2, s. 153-161
-
Tidskriftsartikel (refereegranskat)abstract
- This paper focuses on a method to train a regression model from incomplete input values. It is assumed in this paper that there are no missing values in a training dataset while missing values exist during a prediction phase using the trained model. Under this assumption, we propose Intentional-Value-Substitution (IVS) training to obtain amachine learningmodel that makes the prediction error as minimum as possible. In IVS training, a model is trained to approximate the target function using a modified training dataset in which some feature values are substituted with a certain value even though their values are not missing. It is shown through a series of computational experiments that the substitution values calculated from a mathematical analysis help the models correctly predict outputs for inputs with missing values.
|
|
| 26. |
- Fukushima, Takuya, et al.
(författare)
-
Team classification with tactical analysis using fuzzy inference in robocup soccer
- 2020
-
Ingår i: 2020 Joint 11th International Conference on Soft Computing and Intelligent Systems and 21st International Symposium on Advanced Intelligent Systems (SCIS-ISIS). - : IEEE. - 9781728197326
-
Konferensbidrag (refereegranskat)abstract
- In this paper, we propose an analysis method based on fuzzy inference in order to improve the classification accuracy on team strategies in RoboCup soccer. It is difficult to quantitatively evaluate team strategies because there are no appropriate ways to represent game situations including an intractable number of factors such as field states and tactics. Therefore, the performance of tactical analysis is not high enough to classify unknown teams. Because kick probability distribution proposed in previous works does not consider the kick directions, this paper employs kick direction distribution obtained by kernel density estimation using von-Mises distributions. A fuzzy inference system with the kick probability distribution as well as the kick direction distribution in the antecedent part of the fuzzy rules is constructed on the RoboCup games. This paper evaluates the classification accuracy of the proposed method through a series of computational experiments.
|
|
| 27. |
- Galván, Edgar, et al.
(författare)
-
Agents in a privacy-preserving world
- 2021
-
Ingår i: Transactions on Data Privacy. - : University of Skövde. - 1888-5063 .- 2013-1631. ; 14:1, s. 53-63
-
Tidskriftsartikel (refereegranskat)abstract
- Privacy is a fluid concept. It is both difficult to define and difficult to achieve. The large amounts of data currently available at hands of companies and administrations increase individual concerns on what is yet to be known about us. For the sake of penalisation and customisation, we often need to give up and supply information that we consider sensitive and private. Other sensitive information is inferred from information that seems harmless. Even when we explicitly require privacy and anonymity, profiling and device fingerprinting may disclose information about us leading to reidentification. Mobile devices and the internet of things make keeping our live private still more difficult. Agent technologies can play a fundamental role to provide privacy-aware solutions. Agents are inherently suitable in the heterogeneous environment in which our devices work, and we can delegate to them the task of protecting our privacy. Agents should be able to reason about our privacy requirements, and may collaborate (or not) with other agents to help us to achieve our privacy goals. We are presented in the connected world with multiple interests, profiles, and also through multiple agentified devices. We envision our agentified devices to collaborate among themselves and with other devices so that our privacy preferences are satisfied. We believe that this is an overlooked field. Our work intends to start shedding some light on the topic by outlining the requirements and challenges where agent technologies can provide a decisive role.
|
|
| 28. |
- Garg, Sonakshi, 1998-, et al.
(författare)
-
Can synthetic data preserve manifold properties?
- 2024
-
Ingår i: ICT systems security and privacy protection. - Cham : Springer. - 9783031651748 - 9783031651779 - 9783031651755 ; , s. 134-147
-
Bokkapitel (refereegranskat)abstract
- Machine learning has shown remarkable performance in modeling large datasets with complex patterns. As the amount of data increases, it often leads to high-dimensional feature spaces. This data may contain confidential information that must be safeguarded against disclosure. One way to make the data accessible could be by using anonymization. An alternative is to use synthetic data that mimics the behavior of the original data. GANs represent a prominent approach for generating synthetic samples that faithfully replicate the distributional characteristics of the original data. In scenarios involving high-dimensional data, preserving the geometric properties, structural integrity, and relative positioning of data points is paramount, as neglecting such information may compromise utility. This research aims to investigate the manifold properties of synthetically generated data and introduces a novel framework for producing privacy-preserving synthetic data while upholding the manifold structure of the original data. While existing studies predominantly focus on privacy preservation within GANs, the critical aspect of preserving the manifold structure of data remains unaddressed. Our novel approach adeptly addresses both privacy concerns and manifold structure preservation, distinguishing it from prior research endeavors. Comparative assessments against baseline models are conducted using metrics such as Maximum Mean Discrepancy (MMD), Fréchet Inception Distance (FID), and F1-score. Additionally, the privacy risk posed by the models is evaluated through data reconstruction attacks. Results demonstrate that the proposed framework exhibits diminished vulnerability to privacy breaches while more effectively preserving the intrinsic structure of the data.
|
|
| 29. |
- Garg, Sonakshi, 1998-, et al.
(författare)
-
K-anonymous privacy preserving manifold learning
- 2023
-
Ingår i: Proceedings of the 20th international conference on security and cryptography. - : SciTePress. - 9789897586668 ; , s. 37-48
-
Konferensbidrag (refereegranskat)abstract
- In this modern world of digitalization, abundant amount of data is being generated. This often leads to data of high dimension, making data points far-away from each other. Such data may contain confidential information and must be protected from disclosure. Preserving privacy of this high-dimensional data is still a challenging problem. This paper aims to provide a privacy preserving model to anonymize high-dimensional data maintaining the manifold structure of the data. Manifold Learning hypothesize that real-world data lie on a low-dimensional manifold embedded in a higher-dimensional space. This paper proposes a novel approach that uses geodesic distance in manifold learning methods such as ISOMAP and LLE to preserve the manifold structure on low-dimensional embedding. Later on, anonymization of such sensitive data is achieved by M-MDAV, the manifold version of MDAV using geodesic distance. MDAV is a micro-aggregation privacy model. Finally, to evaluate the efficiency of the prop osed approach machine learning classification is performed on the anonymized lower-embedding. To emphasize the importance of geodesic-manifold learning, we compared our approach with a baseline method in which we try to anonymise high-dimensional data directly without reducing it onto a lower-dimensional space. We evaluate the proposed approach over natural and synthetic data such as tabular, image and textual data sets, and then empirically evaluate the performance of the proposed approach using different evaluation metrics viz. accuracy, precision, recall and K-Stress. We show that our proposed approach is providing accuracy up to 99% and thus, provides a novel contribution of analysing the effects of K-anonymity in manifold learning.
|
|
| 30. |
- Garg, Sonakshi, 1998-, et al.
(författare)
-
Privacy in manifolds : combining k-anonymity with differential privacy on Fréchet means
- 2024
-
Ingår i: Computers & security (Print). - : Elsevier. - 0167-4048 .- 1872-6208. ; 144
-
Tidskriftsartikel (refereegranskat)abstract
- While anonymization techniques have improved greatly in allowing data to be used again, it is still really hard to get useful information from anonymized data without risking people’s privacy. Conventional approaches such as k-Anonymity and Differential Privacy have limitations in preserving data utility and privacy simultaneously, particularly in high-dimensional spaces with manifold structures. We address this challenge by focusing on anonymizing data existing within high-dimensional spaces possessing manifold structures. To tackle these issues, we propose and implement a hybrid anonymization scheme termed as the (?, ?, ?)-anonymization method that combines elements of both differential privacy and k-anonymity. This approach aims to produce high-quality anonymized data that closely resembles real data in terms of knowledge extraction while safeguarding privacy. The Fréchet mean, an operation applicable in metric spaces and meaningful in the manifold setting, serves as a key aspect of our approach. It provides insight into the geometry of data points within high-dimensional spaces. Our goal is to anonymize this Fréchet mean using our proposed approach and minimize the distance between the original and anonymized Fréchet mean to achieve data privacy without significant loss of information. Additionally, we introduce a novel Fréchet mean clustering model designed to enhance the clustering process for high-dimensional spaces. Through theoretical analysis and practical experiments, we demonstrate that our approach outperforms traditional privacy models both in terms of preserving data utility and privacy. This research contributes to advancing privacy-preserving techniques for complex and non-linear data structures, ensuring a balance between data utility and privacy protection.
|
|
| 31. |
- Gupta, Sargam, et al.
(författare)
-
Differentially private traffic flow prediction using transformers : a federated approach
- 2024
-
Ingår i: Computer Security. ESORICS 2023 International Workshops. - : Springer Nature. - 9783031542039 - 9783031542046 ; , s. 260-271
-
Konferensbidrag (refereegranskat)abstract
- Accurate traffic flow prediction plays an important role in intelligent transportation management and reducing traffic congestion for smart cities. Existing traffic flow prediction techniques using deep learning, mostly LSTMs, have achieved enormous success based on the large traffic flow datasets collected by governments and different organizations. Nevertheless, a lot of these datasets contain sensitive attributes that may relate to users’ private data. Hence, there is a need to develop an accurate traffic flow prediction mechanism that preserves users’ privacy. To address this challenge, we propose a federated learning-based temporal fusion transformer framework for traffic flow prediction which is a distributed machine learning approach where all the model updates are aggregated through an aggregation algorithm rather than sharing and storing the raw data in one centralized location. The proposed framework trains the data locally on client devices using temporal fusion transformers and differential privacy. Experiments show that the proposed framework can guarantee accuracy in predicting traffic flow for both the short and long term.
|
|
| 32. |
- Halas, Radomir, et al.
(författare)
-
A note on some algebraic properties of discrete Sugeno integrals
- 2019
-
Ingår i: Fuzzy sets and systems (Print). - : Elsevier. - 0165-0114 .- 1872-6801. ; 355, s. 110-120
-
Tidskriftsartikel (refereegranskat)abstract
- Based on the link between Sugeno integrals and fuzzy measures, we discuss several algebraic properties of discrete Sugeno integrals. We recall that the composition of Sugeno integrals is again a Sugeno integral, and that each Sugeno integral can be obtained as a composition of binary Sugeno integrals. In particular, we discuss the associativity, dominance, commuting and bisymmetry of Sugeno integrals.
|
|
| 33. |
- Hendrick, Noel, et al.
(författare)
-
Genetic Algorithms in Data Masking : Towards Privacy as a Service
- 2021
-
Ingår i: Artificial Intelligence and Soft Computing. - Cham : Springer Nature. - 9783030879853 - 9783030879860 ; , s. 381-391
-
Konferensbidrag (refereegranskat)abstract
- Today’s world is one where the number of publicly stored information and private data is growing exponentially, thus so is the need for more precise and more efficient data protection methods. Data privacy is the field that studies data protection methods as well as privacy models, tools and measures to establish when data is well protected and compliant with privacy requirements. Masking methods are used to perturb a database to permit data analysis while ensuring privacy.This work provides a tool towards privacy as a service. Selecting an appropriate masking method and an appropriate parameterisation is an heuristic process. Our work makes use of genetic algorithms to find a good combination of masking methods and parameters. To do so, a number of solutions (masking methods, parameters) are applied and evaluated, the effectiveness of each solution is measured and well performing solutions are passed on to future generations. Effectiveness of a solution is in terms of information loss and disclosure risk.
|
|
| 34. |
- Jiang, Lili, et al.
(författare)
-
Data protection and multi-database data-driven models
- 2023
-
Ingår i: Future Internet. - : MDPI. - 1999-5903. ; 15:3
-
Tidskriftsartikel (refereegranskat)abstract
- Anonymization and data masking have effects on data-driven models. Different anonymization methods have been developed to provide a good trade-off between privacy guarantees and data utility. Nevertheless, the effects of data protection (e.g., data microaggregation and noise addition) on data integration and on data-driven models (e.g., machine learning models) built from these data are not known. In this paper, we study how data protection affects data integration, and the corresponding effects on the results of machine learning models built from the outcome of the data integration process. The experimental results show that the levels of protection that prevent proper database integration do not affect machine learning models that learn from the integrated database to the same degree. Concretely, our preliminary analysis and experiments show that data protection techniques have a lower level of impact on data integration than on machine learning models.
|
|
| 35. |
- Jiang, Lili, et al.
(författare)
-
On the Effects of Data Protection on Multi-database Data-Driven Models
- 2022
-
Ingår i: Integrated Uncertainty in Knowledge Modelling and Decision Making. - Cham : Springer. - 9783030980177 - 9783030980184 ; , s. 226-238
-
Konferensbidrag (refereegranskat)abstract
- This paper analyses the effects of masking mechanism for privacy preservation in data-driven models (regression) with respect to database integration. Especially two data masking methods (microaggregation and rank swapping) are applied on two public datasets to evaluate the linear regression model in terms of privacy protection and prediction performance. Our preliminary experimental results show that both methods achieve a good trade-off of privacy protection and information loss. We also show that for some experiments although data integration produces some incorrect links, the linear regression model is still comparable, with respect to prediction error, to the one inferred from the original data.
|
|
| 36. |
- Kaya, Sema Kayapinar, et al.
(författare)
-
Dynamic Features Spaces and Machine Learning: Open Problems and Synthetic Data Sets
- 2020
-
Ingår i: Integrated Uncertainty in Knowledge Modelling and Decision Making. - Cham : Springer Nature. - 9783030625085 - 9783030625092 ; , s. 125-136
-
Konferensbidrag (refereegranskat)abstract
- Dynamic feature spaces appear when different records or instances in databases are defined in terms of different features. This is in contrast with usual (static) feature spaces in standard databases, where the schema of the database is known and fixed. Then, all records in the database have the same set of variables, attributes or features. Dynamic feature mining algorithms are to extract knowledge from data on dynamic feature spaces. As an example, spam detection methods have been developed from a dynamic feature space perspective. Words are taken as features and new words appearing in new emails are, therefore, considered new features. In this case, the problem of spam detection is represented as a classification problem (a supervised machine learning problem).The relevance of dynamic feature spaces is increasing. The large amounts of data currently available or received by systems are not necessarily described using the same feature spaces. This is the case of distributed databases with data about customers, providers, etc. Industry 4.0, Internet of Things, and RFIDs are and will be a source of data in dynamic feature spaces. New sensors added in an industrial environment, new devices connected into a smart home, new types of analysis and new types of sensors in healthcare, all are examples of dynamic feature spaces. Machine learning algorithms are needed to deal with these type of scenarios.In this paper we motivate the interest for dynamic feature mining, we give some examples of scenarios where these techniques are needed, we review some of the existing solutions and its relationship with other areas of machine learning and data mining (e.g., incremental learning, concept drift, topic modeling), we discuss some open problems, and we discuss synthetic data generation for this type of problem.
|
|
| 37. |
- Khan, Md Sakib Nizam, 1990-
(författare)
-
Towards Privacy Preserving Intelligent Systems
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Intelligent systems, i.e., digital systems containing smart devices that can gather, analyze, and act in response to the data they collect from their surrounding environment, have progressed from theory to application especially in the last decade, thanks to the recent technological advances in sensors and machine learning. These systems can take decisions on users' behalf dynamically by learning their behavior over time. The number of such smart devices in our surroundings is increasing rapidly. Since these devices in most cases handle privacy-sensitive data, privacy concerns are also increasing at a similar rate. However, privacy research has not been in sync with these developments. Moreover, the systems are heterogeneous in nature (e.g., in terms of form factor, energy, processing power, use case scenarios, etc.) and continuously evolving which makes the privacy problem even more challenging.In this thesis, we identify open privacy problems of intelligent systems and later propose solutions to some of the most prominent ones. We first investigate privacy concerns in the context of data stored on a single smart device. We identify that ownership change of a smart device can leak privacy-sensitive information stored on the device. To solve this, we propose a framework to enhance the privacy of owners during ownership change of smart devices based on context detection and data encryption. Moving from the single-device setting to more complex systems involving multiple devices, we conduct a systematic literature review and a review of commercial systems to identify the unique privacy concerns of home-based health monitoring systems. From the review, we distill a common architecture covering most commercial and academic systems, including an inventory of what concerns they address, their privacy considerations, and how they handle the data. Based on this, we then identify potential privacy intervention points of such systems.For the publication of collected data or a machine-learning model trained on such data, we explore the potential of synthetic data as a tool for achieving a better trade-off between privacy and utility compared to traditional privacy-enhancing approaches. We perform a thorough assessment of the utility of synthetic tabular data. Our investigation reveals that none of the commonly used utility metrics for assessing how well synthetic data corresponds to the original data can predict whether for any given univariate or multivariate statistical analysis (when the analysis is not known beforehand) synthetic data achieves utility similar to the original data. For machine learning-based classification tasks, however, the metric Confidence Interval Overlap shows a strong correlation with how similarly the machine learning models (i.e., trained on synthetic vs. original) perform. Concerning privacy, we explore membership inference attacks against machine learning models which aim at finding out whether some (or someone's) particular data was used to train the model. We find from our exploration that training on synthetic data instead of original data can significantly reduce the effectiveness of membership inference attacks. For image data, we propose a novel methodology to quantify, improve, and tune the privacy utility trade-off of the synthetic image data generation process compared to the traditional approaches.Overall, our exploration in this thesis reveals that there are several open research questions regarding privacy at different phases of the data lifespan of intelligent systems such as privacy-preserving data storage, possible inferences due to data aggregation, and the quantification and improvement of privacy utility trade-off for achieving better utility at an acceptable level of privacy in a data release. The identified privacy concerns and their corresponding solutions presented in this thesis will help the research community to recognize and address remaining privacy concerns in the domain. Solving the concerns will encourage the end-users to adopt the systems and enjoy the benefits without having to worry about privacy.
|
|
| 38. |
- Koloseni, David, et al.
(författare)
-
Absolute and relative preferences in AHP-like matrices
- 2018
-
Ingår i: Data Science and Knowledge Engineering for Sensing Decision Support. - SINGAPORE : World Scientific Publishing Co. Pte. Ltd.. - 9789813273221 - 9789813273245 ; , s. 260-267
-
Konferensbidrag (refereegranskat)abstract
- The Analytical Hierarchy Process (AHP) has been extensively used to interview experts in order to find the weights of the criteria. We call AHP-like matrices relative preferences of weights. In this paper we propose another type of matrix that we call a absolute preference matrix. They are also used to find weights, and we propose that they can be applied to find the weights of weighted means and also of the Choquet integral.
|
|
| 39. |
- Koloseni, David, et al.
(författare)
-
AHP-Like Matrices and Structures : Absolute and Relative Preferences
- 2020
-
Ingår i: Mathematics. - : MDPI. - 2227-7390. ; 8:5
-
Tidskriftsartikel (refereegranskat)abstract
- Aggregation functions are extensively used in decision making processes to combine available information. Arithmetic mean and weighted mean are some of the most used ones. In order to use a weighted mean, we need to define its weights. The Analytical Hierarchy Process (AHP) is a well known technique used to obtain weights based on interviews with experts. From the interviews we define a matrix of pairwise comparisons of the importance of the weights. We call these AHP-like matrices absolute preferences of weights. We propose another type of matrix that we call a relative preference matrix. We define this matrix with the same goal—to find the weights for weighted aggregators. We discuss how it can be used for eliciting the weights for the weighted mean and define a similar approach for the Choquet integral.
|
|
| 40. |
|
|
| 41. |
- Kwatra, Saloni, et al.
(författare)
-
A k-Anonymised Federated Learning Framework with Decision Trees
- 2022
-
Ingår i: Data Privacy Management, Cryptocurrencies and Blockchain Technology. - Cham : Springer Science+Business Media B.V.. - 9783030939434 ; , s. 106-120
-
Konferensbidrag (refereegranskat)abstract
- We propose a privacy-preserving framework using Mondrian k-anonymity with decision trees in a Federated Learning (FL) setting for the horizontally partitioned data. Data heterogeneity in FL makes the data non-IID (Non-Independent and Identically Distributed). We use a novel approach to create non-IID partitions of data by solving an optimization problem. In this work, each device trains a decision tree classifier. Devices share the root node of their trees with the aggregator. The aggregator merges the trees by choosing the most common split attribute and grows the branches based on the split values of the chosen split attribute. This recursive process stops when all the nodes to be merged are leaf nodes. After the merging operation, the aggregator sends the merged decision tree to the distributed devices. Therefore, we aim to build a joint machine learning model based on the data from multiple devices while offering k-anonymity to the participants.
|
|
| 42. |
- Kwatra, Saloni, et al.
(författare)
-
A Survey on Tree Aggregation
- 2021
-
Ingår i: 2021 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). - : IEEE. - 9781665444071 - 9781665444088
-
Konferensbidrag (refereegranskat)abstract
- The research dedicated to the aggregation of classification trees and general trees (hierarchical structure of objects) has made enormous progress in the past decade. The problem statement for aggregation of classification trees or general trees is as follows: Given k classification or general trees for a set of objects, we aim to build a consensus tree (classification or general). That is, a representative tree for the given trees. In this paper, we explore different perspectives for the motivation to construct a single tree from multiple trees given by researchers. The survey presents the approaches for the aggregation of both the classification trees as well as general trees. We bifurcate our study of the aggregation approaches into two categories: Selecting a single tree from multiple trees and merging trees. We will discuss these categories and the aggregation approaches under these categories in the paper comprehensively. We also discuss the privacy aspects of tree aggregation approaches and the possible directions for new research like using the technique of aggregating decision trees in the field of Federated Learning, which is a booming topic.
|
|
| 43. |
- Kwatra, Saloni, et al.
(författare)
-
Data reconstruction attack against principal component analysis
- 2023
-
Ingår i: Security and Privacy in Social Networks and Big Data. - : Springer Science+Business Media B.V.. - 9789819951765 - 9789819951772 ; , s. 79-92
-
Konferensbidrag (refereegranskat)abstract
- Attacking machine learning models is one of the many ways to measure the privacy of machine learning models. Therefore, studying the performance of attacks against machine learning techniques is essential to know whether somebody can share information about machine learning models, and if shared, how much can be shared? In this work, we investigate one of the widely used dimensionality reduction techniques Principal Component Analysis (PCA). We refer to a recent paper that shows how to attack PCA using a Membership Inference Attack (MIA). When using membership inference attacks against PCA, the adversary gets access to some of the principal components and wants to determine if a particular record was used to compute those principal components. We assume that the adversary knows the distribution of training data, which is a reasonable and useful assumption for a membership inference attack. With this assumption, we show that the adversary can make a data reconstruction attack, which is a more severe attack than the membership attack. For a protection mechanism, we propose that the data guardian first generate synthetic data and then compute the principal components. We also compare our proposed approach with Differentially Private Principal Component Analysis (DPPCA). The experimental findings show the degree to which the adversary successfully attempted to recover the users’ original data. We obtained comparable results with DPPCA. The number of principal components the attacker intercepted affects the attack’s outcome. Therefore, our work aims to answer how much information about machine learning models is safe to disclose while protecting users’ privacy.
|
|
| 44. |
- Kwatra, Saloni, et al.
(författare)
-
Empirical evaluation of synthetic data created by generative models via attribute inference attack
- 2024
-
Ingår i: Privacy and identity management. - : Springer. - 9783031579776 - 9783031579783 ; , s. 282-291
-
Konferensbidrag (refereegranskat)abstract
- The disclosure risk of synthetic/artificial data is still being determined. Studies show that synthetic data generation techniques generate similar data to the original data and sometimes even the exact original data. Therefore, publishing synthetic datasets can endanger the privacy of users. In our work, we study the synthetic data generated from different synthetic data generation techniques, including the most recent diffusion models. We perform a disclosure risk assessment of synthetic datasets via an attribute inference attack, in which an attacker has access to a subset of publicly available features and at least one synthesized dataset, and the aim is to infer the sensitive features unknown to the attacker. We also compute the predictive accuracy and F1 score of the random forest classifier trained on several synthetic datasets. For sensitive categorical features, we show that Attribute Inference Attack is not highly feasible or successful. In contrast, for continuous attributes, we can have an approximate inference. This holds true for the synthetic datasets derived from Diffusion models, GANs, and DPGANs, which shows that we can only have approximated Attribute Inference, not the exact Attribute Inference.
|
|
| 45. |
- Kwatra, Saloni, et al.
(författare)
-
Integrally private model selection for support vector machine
- 2024
-
Ingår i: Computer Security. ESORICS 2023 International Workshops. - : Springer Nature. - 9783031542039 - 9783031542046 ; , s. 249-259
-
Konferensbidrag (refereegranskat)abstract
- Today, there are unlimited applications of data mining techniques. According to ongoing privacy regulations, data mining techniques that preserve users’ privacy are a primary requirement. Our work contributes to the Privacy-Preserving Data Mining (PPDM) domain. We work with Integral Privacy, which provides users with private machine learning model recommendations and privacy against model comparison attacks. For machine learning, we work with Support Vector Machine (SVM), which is based on the structural risk minimization principle. Our experiments show that we obtain highly recurrent SVM models due to their peculiar properties, requiring only a subset of the training data to learn well. Not only high recurrence, but from our empirical results, we show that integrally private SVM models obtain good results in accuracy, recall, precision, and F1-score compared with the baseline SVM model and the ϵ Differentially Private SVM (DPSVM) model.
|
|
| 46. |
- Minh-Ha, Le, 1989-
(författare)
-
Beyond Recognition : Privacy Protections in a Surveilled World
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- This thesis addresses the need to balance the use of facial recognition systems with the need to protect personal privacy in machine learning and biometric identification. As advances in deep learning accelerate their evolution, facial recognition systems enhance security capabilities, but also risk invading personal privacy. Our research identifies and addresses critical vulnerabilities inherent in facial recognition systems, and proposes innovative privacy-enhancing technologies that anonymize facial data while maintaining its utility for legitimate applications.Our investigation centers on the development of methodologies and frameworks that achieve k-anonymity in facial datasets; leverage identity disentanglement to facilitate anonymization; exploit the vulnerabilities of facial recognition systems to underscore their limitations; and implement practical defenses against unauthorized recognition systems. We introduce novel contributions such as AnonFACES, StyleID, IdDecoder, StyleAdv, and DiffPrivate, each designed to protect facial privacy through advanced adversarial machine learning techniques and generative models. These solutions not only demonstrate the feasibility of protecting facial privacy in an increasingly surveilled world, but also highlight the ongoing need for robust countermeasures against the ever-evolving capabilities of facial recognition technology.Continuous innovation in privacy-enhancing technologies is required to safeguard individuals from the pervasive reach of digital surveillance and protect their fundamental right to privacy. By providing open-source, publicly available tools, and frameworks, this thesis contributes to the collective effort to ensure that advancements in facial recognition serve the public good without compromising individual rights. Our multi-disciplinary approach bridges the gap between biometric systems, adversarial machine learning, and generative modeling to pave the way for future research in the domain and support AI innovation where technological advancement and privacy are balanced.
|
|
| 47. |
|
|
| 48. |
|
|
| 49. |
-
Modeling Decisions for Artificial Intelligence : 18th International Conference, MDAI 2021, Umeå, Sweden, September 27–30, 2021, Proceedings
- 2021
-
Proceedings (redaktörskap) (refereegranskat)abstract
- This book constitutes the refereed proceedings of the 18th International Conference on Modeling Decisions for Artificial Intelligence, MDAI 2021, held in Umeå, Sweden, in September 2021.*The 24 papers presented in this volume were carefully reviewed and selected from 50 submissions. Additionally, 3 invited papers were included. The papers discuss different facets of decision processes in a broad sense and present research in data science, data privacy, aggregation functions, human decision making, graphs and social networks, and recommendation and search. The papers are organized in the following topical sections: aggregation operators and decision making; approximate reasoning; machine learning; data science and data privacy.*The conference was held virtually due to the COVID-19 pandemic.
|
|
| 50. |
|
|
