Sensitive Attributes Research Articles

Learning Dynamic Prototypes for Visual Pattern Debiasing

AbstractDeep learning has achieved great success in academic benchmarks but fails to work effectively in the real world due to the potential dataset bias. The current learning methods are prone to inheriting or even amplifying the bias present in a training dataset and under-represent specific demographic groups. More recently, some dataset debiasing methods have been developed to address the above challenges based on the awareness of protected or sensitive attribute labels. However, the number of protected or sensitive attributes may be considerably large, making it laborious and costly to acquire sufficient manual annotation. To this end, we propose a prototype-based network to dynamically balance the learning of different subgroups for a given dataset. First, an object pattern embedding mechanism is presented to make the network focus on the foreground region. Then we design a prototype learning method to discover and extract the visual patterns from the training data in an unsupervised way. The number of prototypes is dynamic depending on the pattern structure of the feature space. We evaluate the proposed prototype-based network on three widely used polyp segmentation datasets with abundant qualitative and quantitative experiments. Experimental results show that our proposed method outperforms the CNN-based and transformer-based state-of-the-art methods in terms of both effectiveness and fairness metrics. Moreover, extensive ablation studies are conducted to show the effectiveness of each proposed component and various parameter values. Lastly, we analyze how the number of prototypes grows during the training process and visualize the associated subgroups for each learned prototype. The code and data will be released at https://github.com/zijinY/dynamic-prototype-debiasing.

FAL-CUR: Fair Active Learning using Uncertainty and Representativeness on Fair Clustering

Active Learning (AL) techniques have proven to be highly effective in reducing data labeling costs across a range of machine learning tasks. Nevertheless, one known challenge of these methods is their potential to introduce unfairness towards sensitive attributes. Although recent approaches have focused on enhancing fairness in AL, they tend to reduce the model’s accuracy. To address this issue, we propose a novel strategy, named Fair Active Learning using fair Clustering, Uncertainty, and Representativeness (FAL-CUR), to improve fairness in AL. FAL-CUR tackles the fairness problem in AL by combining fair clustering with an acquisition function that determines which samples to query based on their uncertainty and representativeness scores. We evaluate the performance of FAL-CUR on four real-world datasets, and the results demonstrate that FAL-CUR achieves a 15%–20% improvement in fairness compared to the best state-of-the-art method in terms of equalized odds while maintaining stable accuracy scores. Furthermore, an ablation study highlights the crucial roles of fair clustering in preserving fairness and the acquisition function in stabilizing the accuracy performance.

Analysis of Factors Leveraging Sustainability of Community-Based Drinking Water Supply (Case Study of Drinking Water Supply for The Pamsimas Program, Kapongan Sub-District, Situbondo Regency)

This research aims to identify the sustainability driving factors of drinking water provision in the PAMSIMAS program in Kapongan Sub-District, Situbondo Regency, focusing on community participation based on environmental, social, economic, technological, and institutional dimensions. This study utilizes a quantitative approach with the Rap-SPAM analysis method, a modification of Rapfish. The data used consist of both primary and secondary sources. Primary data are obtained through surveys, observations, and interviews with respondents, utilizing questionnaires. Secondary data is gathered through relevant literature reviews and various related institutions. The results of the Rap-SPAM simulation on 5 dimensions of sustainability, obtained 18 sensitive attributes or main leveraging factors. The main leverage attributes in the environmental dimension are the surrounding land condition, pollution potential, rainfall & rainy days, and quantity of raw water. In the social dimension, the main leverage attributes are community meetings, practices of using the water supply system, and community trust. In the economic dimension, the main leverage attributes are existence of fees for the development of the water supply system network, availability of funds for the development of the water supply system, existence of user fees, and affordability of user fees. In the technological dimension, the main leverage attributes are water efficiency of water use, ease of technology application, network leakages, and quality of received water. In the institutional dimension, the main leverage attributes are clear organizational structure, performance and financial reporting by the management to the users of the water supply system, and regulations for water supply management.

An Access Control Framework for Multilayer Rail Transit Systems Based on Trust and Sensitivity Attributes

The construction of multilayer rail transit systems is a necessary way to realize “modern metropolitan areas on rail”, improve resource sharing, and increase travel services, where data integration is of utmost importance. To break data silos and realize data flow between different rail systems, a fine-grained access control framework is proposed in this paper. Through categorical and hierarchical schemes, a universal security scale is established for cross-domain data resources. Based on this, a trust and sensitivity attribute-based access control (TSABAC) model is put forward to describe the characteristics of the access control process. Furthermore, the method of policy integration is discussed, as well as the solution to the policy incompatibility problem, due to cross-domain interaction. As shown in practical application and simulation analysis, this framework can meet the requirements of security and granularity. This research is of great significance for promoting the high-quality development of urban agglomerations and metropolitan areas, and improving the quality and efficiency of rail transit.

Profit vs. Equality? The Case of Financial Risk Assessment and A New Perspective on Alternative Data

The importance of pursuing financial inclusion to accelerate economic growth and enhance financial sustainability has been well noted. However, studies have provided few actionable insights into how financial institutions can balance the potential socioeconomic trade-off between profitability and equality. One major challenge arises from a lack of understanding of the impacts of various types of market information available on financial equality beyond economic profitability. Another challenge lies in how the socioeconomic trade-off under a large set of counterfactual policies in a real-world setting can be evaluated. Our motivation for the present study was the emerging sources of digitized user-behavior data (i.e., “alternative data”) stemming from the high penetration of mobile devices and internet access. Accordingly, we investigated how alternative data from smartphones and social media can help mitigate potential financial inequality while preserving business profitability in the context of financial credit risk assessment. We partnered with a leading microloan website to design a novel “meta” experiment that allowed us to simulate various real-world field experiments under an exhaustive set of counterfactual policies. Interestingly, we found that profiling user financial risk using smartphone activities is 1.3 times more effective in improving financial inclusion than using online social media information (23.05% better vs. 18.11%), and nearly 1.3 times more effective in improving business profitability (42% better vs. 33%). Surprisingly, we found that using consumers’ online shopping activities for credit risk profiling can hurt financial inclusion. Furthermore, we investigated potential explanations for financial inclusion improvements. Our findings suggest that alternative data, especially users’ smartphone activities, not only demonstrate higher ubiquity but also appear to be more orthogonal to conventional sensitive demographic attributes. This, in turn, can help mitigate statistical bias driven by the unobserved factors or underrepresentative training samples in machine-based risk assessment processes.

A transformer-based deep learning approach for fairly predicting post-liver transplant risk factors

Liver transplantation is a life-saving procedure for patients with end-stage liver disease. There are two main challenges in liver transplant: finding the best matching patient for a donor and ensuring transplant equity among different subpopulations. The current MELD scoring system evaluates a patient's mortality risk if not receiving an organ within 90 days. However, the donor-patient matching should also consider post-transplant risk factors, such as cardiovascular disease, chronic rejection, etc., which are all common complications after transplant. Accurate prediction of these risk scores remains a significant challenge. In this study, we used predictive models to solve the above challenges. Specifically, we proposed a deep learning model to predict multiple risk factors after a liver transplant. By formulating it as a multi-task learning problem, the proposed deep neural network was trained to simultaneously predict the five post-transplant risks and achieve equal good performance by exploiting task-balancing techniques. We also proposed a novel fairness-achieving algorithm to ensure prediction fairness across different subpopulations. We used electronic health records of 160,360 liver transplant patients, including demographic information, clinical variables, and laboratory values, collected from the liver transplant records of the United States from 1987 to 2018. The model's performance was evaluated using various performance metrics such as AUROC and AUPRC. Our experiment results highlighted the success of our multi-task model in achieving task balance while maintaining accuracy. The model significantly reduced the task discrepancy by 39 %. Further application of the fairness-achieving algorithm substantially reduced fairness disparity among all sensitive attributes (gender, age group, and race/ethnicity) in each risk factor. It underlined the potency of integrating fairness considerations into the task-balancing framework, ensuring robust and fair predictions across multiple tasks and diverse demographic groups.

Learning fair representations via rebalancing graph structure

Graph Neural Network (GNN) models have been extensively researched and utilised for extracting valuable insights from graph data. The performance of fairness algorithms based on GNNs depends on the neighbourhood aggregation mechanism during the update process. However, this mechanism may result in the disregard of sensitive attributes pertaining to nodes with low degrees, as well as the disproportionate influence of sensitive attributes associated with high-degree nodes on their neighbouring nodes. To address these limitations, we propose a novel algorithm called Structural Rebalancing Graph Neural Network (SRGNN). SRGNN aims to consider the impact of both low-degree and high-degree nodes in the GNN model for learning fair representations in decision-making. SRGNN first proposes a fair structural rebalancing algorithm to ensure equal status among nodes by reducing the influence of high-degree nodes and enhancing the influence of low-degree nodes. Next, SRGNN utilises adversarial learning to learn fair representations, based on gradient normalisation to ensure that each node’s representation is separated from sensitive attribute information. We conducted extensive experiments on three real-world datasets to evaluate the performance of SRGNN. The results showed that SRGNN outperformed all existing models in 2 out of 2 fairness metrics.

Ensuring generalized fairness in batch classification

In this paper, we consider the problem of batch classification and propose a novel framework for achieving fairness in such settings. The problem of batch classification involves selection of a set of individuals, often encountered in real-world scenarios such as job recruitment, college admissions etc. This is in contrast to a typical classification problem, where each candidate in the test set is considered separately and independently. In such scenarios, achieving the same acceptance rate (i.e., probability of the classifier assigning positive class) for each group (membership determined by the value of sensitive attributes such as gender, race etc.) is often not desirable, and the regulatory body specifies a different acceptance rate for each group. The existing fairness enhancing methods do not allow for such specifications and hence are unsuited for such scenarios. In this paper, we define a configuration model whereby the acceptance rate of each group can be regulated and further introduce a novel batch-wise fairness post-processing framework using the classifier confidence-scores. We deploy our framework across four real-world datasets and two popular notions of fairness, namely demographic parity and equalized odds. In addition to consistent performance improvements over the competing baselines, the proposed framework allows flexibility and significant speed-up. It can also seamlessly incorporate multiple overlapping sensitive attributes. To further demonstrate the generalizability of our framework, we deploy it to the problem of fair gerrymandering where it achieves a better fairness-accuracy trade-off than the existing baseline method.

Efficient Feature Selection in High Dimensional Data Based on Enhanced Binary Chimp Optimization Algorithms and Machine Learning

Feature selection with the highest performance accuracy is the biggest win for multidimensional data. The Chimpanzee Optimization Algorithm (ChOA) serves as a crucial technique for dealing with multidimensional global optimization issues. However, ChOA often lacks fast convergence and good selection of sensitive attributes leading to poor performance. To address these issues, most significant features were selected using two variants of ChOA called BChimp1 and BChimp2 (BChimp1 and BChimp are available at : https://www.mathworks.com/matlabcentral/fileexchange/133267-binary-chimpoptimization-algorithm-forfeatures-selection. September 22, 202). BChimp1 selects the optimal solution from the four best possible solutions and it applies a stochastic crossover on four moving solutions to deeply speed-up convergence level. BChimp2 uses the sigmoid function to select the significant features. Then, these features were trained using six-well known classifiers. The proposed techniques tend to select the most significant features, speed up the convergence rate and decrease training time for high-dimensional data. 23 standard datasets with six well-known classifiers were employed to assess the performance of BChimp1 and BChimp2. Experimental results validate the efficiency of BChimp1 and BChimp2 in enhancing accuracy by 83.83% and 82.02%, and reducing dimensionality by 42.77% and 72.54%, respectively. However, time-evaluation results of BChimp1 and BChimp2 in all datasets showed fast convergence and surpassed current optimization algorithms such as PSO, GWA, GOA, and GA.

Unawareness detection: Discovering black-box malicious models and quantifying privacy leakage risks

Because machine learning models, especially black-box malicious models vulnerable to attribute inference attacks, are capable of generating a great deal of privacy leakage, recent work has focused on assessing these models in an attempt to prevent unexpected attribute privacy leakage. While there has been some success at model privacy risk evaluations, these traditional solutions are almost brittle in practice because they not only require white-box access to obtain model feature layer outputs but also their evaluation results are heavily influenced by the training dataset and the model structure, leading to difficulty in generalization. In this paper, we propose a novel unawareness detection mechanism for discovering black-box malicious models and quantifying potential unawareness privacy leakage risk along with machine learning models to overcome the two limitations. A new method for quantifying the privacy risk caused by a specific loss function has been proposed to mitigate the impact of the training dataset and the model structure. A new evaluation model has also been proposed that uses Matthew's correlation coefficient score as a new metric and the final output of the target model as a new input. In addition, the theoretical upper bound of the model privacy risk has also been given a mathematical formula that is positively correlated to the mutual information between the sensitive attributes and the target model outputs. Compared with traditional detection methods, our evaluation model reduces the requirement for model access and minimizes evaluation errors caused by data imbalance, and our privacy risk assessment method and theoretical upper bounds on privacy risk can be applied to a broader range of datasets and target model structures. The experimental results show that the adversary's prediction capability is affected by the distribution of datasets and the level of malicious intent of the model, which is consistent with the theoretical prediction, and the detecting method can find potential model privacy leaks in public datasets UTKFace and FairFace.

Trapped in the search box: An examination of algorithmic bias in search engine autocomplete predictions

This paper examines the autocomplete algorithmic bias of leading search engines against three sensitive attributes: gender, race, and sexual orientation. By simulating search query prefixes and calling search engine APIs, 106,896 autocomplete predictions were collected, and their semantic toxicity scores as measures of negative algorithmic bias were computed based on machine learning models. Results indicate that search engine autocomplete algorithmic bias is overall consistent with long-standing societal discrimination. Historically disadvantaged groups such as the female, the Black, and the homosexual suffer higher levels of negative algorithmic bias. Moreover, the degree of algorithmic bias varies across topic categories. Implications about the search engine mediatization, mechanisms and consequences of autocomplete algorithmic bias are discussed.

Sustainability Study of Horticultural Development in Mandailing Natal District With RAPFISH-MDS Analysis

Mandailing Natal district as the second largest district in North Sumatra Province has great potential for horticultural development. This study aimed to determine the sustainability level of horticultural development in Mandailing Natal district through 5 dimensions, namely ecological, economic, social, technological and institutional. The RAPFISH-MDS analysis method was used to measure the index and sustainability status. The results showed that the criteria for the sustainability of horticultural development in Mandailing Natal district was quite sustainable (52.61) with the sustainability index value of the ecological, institutional, social, economic and technological dimensions, respectively, 58.06; 56.86; 53.76; 51.99; and 51.08. Sensitive attributes supporting sustainability are erosion rate for the ecological dimension, post harvest management for the economic dimension, extension intensity for the social dimension, land management technology for the technological dimension, and conflicts between farmer groups for the institutional dimension. The realization of horticultural development can be through the development of government programs and extensions that change farmers' perceptions of the environment and the application of eco friendly technologies in their farming.

Sustainability Analysis of Independent Palm Oil Plantations in Sintang Regency, West Kalimantan

The goal of this investigation is to examine the impact of institutional, technological, social, and economic factors on the long-term viability of independent oil palm plantations in West Kalimantan's Sintang Regency. The data obtained from interviewing 216 respondents, spread across three sub-districts, namely, the Sintang, Kelam Permai, and Binjai Hulu; and used Multi-dimensional Scaling RAPFISH (Rapid Assessment Techniques for Fisheries) to analyze the data. The research results show that the sustainability index values in different aspects are generally favorable, such as the sustainability index value in the economic aspect is 74.36%, the value in the social aspect is 71.28%, the environmental value is 70.59%, the institutional dimension value is 72.34% and the technology dimension value is 70.14%. Considering the sensitive attributes of this research as decisive factors some efforts should be made to sustained the oil palm plantations. Furthermore, governments and policymakers should strengthen their institutional and organizational role in supporting oil palm plantations to make it easier for farmers to access technology, assistance with plantation facilities, and agricultural extension services.

Towards comparable ratings: Exploring bias in German physician reviews

In this study, we evaluate the impact of gender-biased data from German-language physician reviews on the fairness of fine-tuned language models. For two different downstream tasks, we use data reported to be gender biased and aggregate it with annotations. First, we propose a new approach to aspect-based sentiment analysis that allows identifying, extracting, and classifying implicit and explicit aspect phrases and their polarity within a single model. The second task we present is grade prediction, where we predict the overall grade of a review on the basis of the review text. For both tasks, we train numerous transformer models and evaluate their performance. The aggregation of sensitive attributes, such as a physician’s gender and migration background, with individual text reviews allows us to measure the performance of the models with respect to these sensitive groups. These group-wise performance measures act as extrinsic bias measures for our downstream tasks. In addition, we translate several gender-specific templates of the intrinsic bias metrics into the German language and evaluate our fine-tuned models. Based on this set of tasks, fine-tuned models, and intrinsic and extrinsic bias measures, we perform correlation analyses between intrinsic and extrinsic bias measures. In terms of sensitive groups and effect sizes, our bias measure results show different directions. Furthermore, correlations between measures of intrinsic and extrinsic bias can be observed in different directions. This leads us to conclude that gender-biased data does not inherently lead to biased models. Other variables, such as template dependency for intrinsic measures and label distribution in the data, must be taken into account as they strongly influence the metric results. Therefore, we suggest that metrics and templates should be chosen according to the given task and the biases to be assessed.

Characterization and Classification of Soils in Agroecosystems in a Moist Enclave in Northeastern Brazil

Modern times have required studies that take into account the main soil features, aligning the use of land with the protection of more sensitive environments. From this perspective, this study aimed to perform a morphological description and determine the physical and chemical attributes for soil classification in the community of Poção, located in the municipality of Martins/RN, by highlighting the more sensitive attributes in the differentiation of environments through multivariate analysis. Nine soil profiles were identified to perform the morphological description and collect samples for physical and chemical analyses. The study updates the soil classes found in the study area: Acrisols, Planosols, and Cambisols, with the relief being the main factor responsible for the difference between soil attributes. The influence of organic matter on the soil attributes highlights the importance of its maintenance. Aluminum and the clay fraction are responsible for the distinction of the Acrisol class, whereas silt, potassium sodium, total organic carbon, the electrical conductivity of the saturation extract, and the cation exchange capacity allow the differentiation of Fluvisols.

TestSGD : Interpretable Testing of Neural Networks against Subtle Group Discrimination

Discrimination has been shown in many machine learning applications, which calls for sufficient fairness testing before their deployment in ethic-relevant domains. One widely concerning type of discrimination, testing against group discrimination, mostly hidden , is much less studied, compared with identifying individual discrimination . In this work, we propose TestSGD , an interpretable testing approach that systematically identifies and measures hidden (which we call “subtle”) group discrimination of a neural network characterized by conditions over combinations of the sensitive attributes . Specifically, given a neural network, TestSGD first automatically generates an interpretable rule set that categorizes the input space into two groups. Alongside, TestSGD also provides an estimated group discrimination score based on sampling the input space to measure the degree of the identified subtle group discrimination, which is guaranteed to be accurate up to an error bound. We evaluate TestSGD on multiple neural network models trained on popular datasets including both structured data and text data. The experiment results show that TestSGD is effective and efficient in identifying and measuring such subtle group discrimination that has never been revealed before. Furthermore, we show that the testing results of TestSGD can be used to mitigate such discrimination through retraining with negligible accuracy drop.

Sustainability Sustainability Status Index of Simple Agroforestry of Magersaren (SAM) with RAP-SAS Method

The simple agroforestry system of the Magersaren in UB Forest is a source of livelihood for the community that has long inhabited the area. Unsustainable management will increase the amount of damage. So, it needs information about the sustainability status of the simple agroforestry system of the magersaren on various dimensions with sustainable attribute criteria. This research was carried out in the UB Forest educational forest located on the slopes of Mount Arjuno, Sumbersari Hamlet, Tawang Argo Village, Karangploso District, Malang Regency. This study aims to analyze the sustainability status of the agroforestry system of the Magersaren in the UB Forest Area. Sustainability status is determined based on a sustainability index formulated through multi-dimensional scaling (MDS) analysis of the RAP-SAS (Rapid Appraisal for Simple Agroforestry System) method. Based on the results of the sustainability status analysis, it shows that the simple agroforestry system in UB Forest is good or very sustainable, in the ecological dimension with an index value of 85.16%, the economic dimension of 79.70% and the socio-institutional dimension of 72.91%. Meanwhile, the management and technology dimension were 66.94% which showed a fairly sustainable status with sensitive attributes owned, namely business planning and processing with an RMS (Root Mean Square) value of 12.3.

Learning fair models without sensitive attributes: A generative approach

Most existing fair classifiers rely on sensitive attributes to achieve fairness. However, for many scenarios, we cannot obtain sensitive attributes due to privacy and legal issues. The lack of sensitive attributes challenges many existing fair classifiers. Though we lack sensitive attributes, for many applications, there usually exists features/information of various formats that are relevant to sensitive attributes. For example, a person’s purchase history can reflect his/her race, which would help for learning fair classifiers on race. However, the work on exploring relevant features for learning fair models without sensitive attributes is rather limited. Therefore, in this paper, we study a novel problem of learning fair models without sensitive attributes by exploring relevant features. We propose a probabilistic generative framework to effectively estimate the sensitive attribute from the training data with relevant features in various formats and utilize the estimated sensitive attribute information to learn fair models. Experimental results on real-world datasets show the effectiveness of our framework in terms of both accuracy and fairness. Our source code is available at: https://github.com/huaishengzhu/FairWS.

Source Rock Evaluation of Hydrocarbons in Deep-Water Offshore Areas Based on a BP Neural Network

Due to the lack of drilling data and poor quality of seismic data in deep-water offshore areas, conventional methods cannot effectively predict the total organic carbon (TOC) content. In this paper, the BP neural network method is used to predict the TOC of the strata overlying the target layer, which adds to the TOC information in the study area. Then, the highest TOC value of the strata overlying the target layer is used to select the most sensitive seismic attributes. Finally, the sensitive seismic attributes are used to evaluate the source rocks with no or few wells. A set of TOC prediction technology flows is established for TOC combined with seismic attributes under the condition of no wells and few wells in deep-water areas. The application example shows the reliability of TOC prediction by this technical process, and the study has a certain reference significance for the evaluation of hydrocarbon source rocks in offshore deep water.

The improvement of horticultural villages sustainability in Central Java Province, Indonesia

Abstract The potential for vegetables in Indonesia is enormous. However, activities both upstream and downstream are still not well developed so that Indonesia is still importing horticulture to meet domestic needs. Because the potential for horticulture has not been maximized, the Ministry of Agriculture is holding a Horticultural Village program, one of which is in Central Java. The Horticultural Village Program that has been running requires a strategy to increase its sustainability. This research was conducted in Central Java to determine strategies to improve the sustainability of the horticultural village program. The method used in this study is multidimensional scaling (MDS) with the rapid appraisal for horticultural village (RAP-HORVI) technique or RAP-HORVIs. This research method will produce an index value that will determine the sustainability status of all dimensions used, namely, input, process, output, and outcome. The result of this analysis is that the Horticultural Village in Central Java is in a quite sustainable status. The results of this MDS show that 12 sensitive attributes can be used as strategies in increasing the value and sustainability status of Horticultural Villages in Central Java, namely, business management training, availability of production tools and materials, suitability of production input facilities, technical training, availability of capital, optimizing the use of infrastructure assistance, checking on farmers/micro, small, and medium enterprises beneficiaries of SK CPCL, intensive assistance by trainers, increasing planting area, increasing the number and variety of plant types, planning the area of land ready for planting, and increasing business capital.