Fairness-aware learning is increasingly important in data mining. Discrimination prevention aims to prevent discrimination the training before it used conduct predictive analysis. In this paper, we focus on fair generation that ensures generated free. Inspired by generative adversarial networks (GAN), present fairness-aware networks, called FairGAN, which are able learn a generator producing and also preserving good utility. Compared with naive models, FairGAN further classifiers trained can...

In this paper, we investigate the problem of discovering both direct and indirect discrimination from historical data, removing discriminatory effects before data is used for predictive analysis (e.g., building classifiers). The main drawback existing methods that they cannot distinguish part influence really caused by all correlated influences. our approach, make use causal network to capture structure data. Then model as path-specific effects, which accurately identify two types...

Based on imbalanced data, the predictive models for 5-year survivability of breast cancer using decision tree are proposed. After data preprocessing from SEER datasets, it is obviously that category distribution imbalanced. Under-sampling taken to make up disadvantage performance caused by data. The evaluated AUC under ROC curve, accuracy, specificity and sensitivity with 10-fold stratified cross-validation. best while approximately equal. Bagging algorithm used build an integration model...

Fairness-aware learning studies the problem of building machine models that are subject to fairness requirements. Counterfactual is a notion derived from Pearl's causal model, which considers model fair if for particular individual or group its prediction in real world same as counterfactual where individual(s) had belonged different demographic group. However, an inherent limitation it cannot be uniquely quantified observational data certain situations, due unidentifiability quantity. In...

Achieving fairness in learning models is currently an imperative task machine learning. Meanwhile, recent research showed that should be studied from the causal perspective, and proposed a number of criteria based on Pearl's modeling framework. In this paper, we investigate problem building fairness-aware generative adversarial networks (CFGAN), which can learn close distribution given dataset, while also ensuring various graph. CFGAN adopts two generators, whose structures are purposefully...

Discrimination discovery and prevention/removal are increasingly important tasks in data mining. aims to unveil discriminatory practices on the protected attribute (e.g., gender) by analyzing dataset of historical decision records, discrimination prevention remove modifying biased before conducting predictive analysis. In this paper, we show that key is find meaningful partitions can be used provide quantitative evidences for judgment discrimination. With support causal graph, present a...

Anti-discrimination is an increasingly important task in data science. In this paper, we investigate the problem of discovering both direct and indirect discrimination from historical data, removing discriminatory effects before are used for predictive analysis (e.g., building classifiers). The main drawback existing methods that they cannot distinguish part influence really caused by all correlated influences. our approach, make use causal graph to capture structure data. Then, model as...

A recent trend of fair machine learning is to define fairness as causality-based notions which concern the causal connection between protected attributes and decisions. However, one common challenge all identifiability, i.e., whether they can be uniquely measured from observational data, a critical barrier applying these real-world situations. In this paper, we develop framework for measuring different fairness. We propose unified definition that covers most previous notions, namely...

Most of the spacecraft telemetry anomaly detection methods based on statistical models suffer from problems high false negatives, long time consumption, and poor interpretability. Besides, complex interactions, which may determine propagation anomalous mode between parameters, are often ignored. To discover interaction parameters improve efficiency accuracy detection, we propose an framework parametric causality Double-Criteria Drift Streaming Peaks Over Threshold (DCDSPOT). We Normalized...

Predictive models learned from historical data are widely used to help companies and organizations make decisions. However, they may digitally unfairly treat unwanted groups, raising concerns about fairness discrimination. In this paper, we study the fairness-aware ranking problem which aims discover discrimination in ranked datasets reconstruct fair ranking. Existing methods mainly based on statistical parity that cannot measure true discriminatory effect since is causal. On other hand,...

In this paper, we study the fairness-aware classification problem by formulating it as a constrained optimization problem. Several limitations exist in previous works due to lack of theoretical framework for guiding formulation. We propose general address limitations. Our provides: (1) various fairness metrics that can be incorporated into classic models constraints; (2) convex solved efficiently; and (3) lower upper bounds real-world measures are established using surrogate functions,...

Automatic heuristic design through reinforcement learning opens a promising direction for solving computationally difficult problems. Unlike most previous works that aimed at solution construction, we explore the possibility of acquiring local search heuristics massive experiments. To illustrate applicability, an agent is trained to perform walk in space by selecting candidate neighbor each step. Specifically, target floorplanning problem, where generated perturbing sequence pair encoding...

Investigating causality to establish novel criteria for training robust natural language processing (NLP) models is an active research area. However, current methods face various challenges such as the difficulties in identifying keyword lexicons and obtaining data from multiple labeled environments. In this paper, we study problem of NLP a complementary but different angle: treat behavior attack model complex causal mechanism quantify its algorithmic information using minimum description...

In discrimination-aware classification, the pre-process methods for constructing a discrimination-free classifier first remove discrimination from training data, and then learn cleaned data. However, they lack theoretical guarantee potential when is deployed prediction. this paper, we fill gap by mathematically bounding in We adopt causal model modeling data generation mechanism, formally defining population, dataset, obtain two important results: (1) prediction can still exist even if...

Multi-mode NOON states have been attracting increasing attentions recently for their abilities of obtaining supersensitive and superresolved measurements simultaneous multiple-phase estimation. In this paper, four different methods generating multi-mode with a high photon number were proposed. The first method is linear optical approach that makes use the Fock state filtration to reduce lower-order terms from coherent inputs, which are jointly combined produce triggering multi-fold...

Fair machine learning is receiving an increasing attention in fields. Researchers fair have developed correlation or association-based measures such as demographic disparity, mistreatment calibration, causal-based total effect, direct and indirect discrimination, counterfactual fairness, fairness notions equality of opportunity equalized odds that consider both decisions the training data made by predictive models. In this paper, we develop a new notation, called effort. Different from...

In online recommendation, customers arrive in a sequential and stochastic manner from an underlying distribution the decision model recommends chosen item for each arriving individual based on some strategy. We study how to recommend at step maximize expected reward while achieving user-side fairness customers, i.e., who share similar profiles will receive regardless of their sensitive attributes items being recommended. By incorporating causal inference into bandits adopting soft...

In the urban transportation system, unbalanced relationship between taxi demand and number of running taxis reduces drivers' income levels passengers' satisfaction. With help vehicular global positioning system (GPS) data, distribution city can be analyzed to provide advice for drivers. A clustering algorithm called Density-Based Spatial Clustering Applications with Noise (DBSCAN) is suitable discovering hotspots. However, execution efficiency still a big challenge when DBSCAN applied on...