arXiv subjects

Mostafa Milani

Publications and source records attributed to Mostafa Milani.

At least 19 recordsLinked to original sources

arXiv2026

Benchmarking Instance-Dependent Label Noise with Controlled Corruptions

Synthetic instance-dependent label noise (IDN) benchmarks are widely used to evaluate noisy-label learning methods, yet existing approaches typically generate noise through imperfect annotators or classifier raters, leaving the source of ambiguity implicit. We introduce CILN, a benchmark generation framework that creates IDN through controlled input corruptions. A diverse voter pool labels corrupted instances, producing benchmark datasets in which both the source and severity of ambiguity are explicit and controllable. Using CIFAR10, MNIST, and Adult, we construct 90 benchmark settings spanning multiple corruption families and severity levels. Our experiments show that the resulting benchmarks exhibit genuine instance-dependent noise, provide diverse confusion structures, and, on CIFAR-10, can produce label distributions that are closer to human uncertainty than an existing synthetic IDN benchmark. We further demonstrate that corruption-mediated IDN can expose failure modes of popular noisy-label learning methods, including Co-Teaching and DivideMix, that are not observed under comparable levels of rater-fallibility noise. These findings suggest that noise structure, not only noise rate, plays an important role in benchmark difficulty and algorithm behavior. By making ambiguity generation explicit and controllable, CILN provides a complementary benchmarking framework for studying noisy-label learning under diverse sources of instance difficulty.

cs.LG↗

arXiv2026

Structure Selection for Fairness-Constrained Differentially Private Data Synthesis

Differential privacy (DP) enables safe data release, with synthetic data generation emerging as a common approach in recent years. Yet standard synthesizers preserve all dependencies in the data, including spurious correlations between sensitive attributes and outcomes. In fairness-critical settings, this reproduces unwanted bias. A principled remedy is to enforce conditional independence (CI) constraints, which encode domain knowledge or legal requirements that outcomes be independent of sensitive attributes once admissible factors are accounted for. DP synthesis typically proceeds in two phases: (i) a measure- ment step that privatizes selected marginals, often structured via maximum spanning trees (MSTs), and (ii) a reconstruction step that fits a probabilistic model consistent with the noisy marginals. We propose PrivCI, which enforces CI during the measurement step via a CI-aware greedy MST algorithm that integrates feasibility checks into Kruskal's construction under the exponential mechanism, improving accuracy over competing methods. Experiments on standard fairness benchmarks show that PrivCI achieves stronger fidelity and predictive accuracy than prior baselines while satisfying the specified CI constraints.

cs.DB↗

arXiv2026

The Human Factor in Data Cleaning: Exploring Preferences and Biases

Data cleaning is often framed as a technical preprocessing step, yet in practice it relies heavily on human judgment. We report results from a controlled survey study in which participants performed error detection, data repair and imputation, and entity matching tasks on census-inspired scenarios with known semantic validity. We find systematic evidence for several cognitive bias mechanisms in data cleaning. Framing effects arise when surface-level formatting differences (e.g., capitalization or numeric presentation) increase false-positive error flags despite unchanged semantics. Anchoring and adjustment bias appears when expert cues shift participant decisions beyond parity, consistent with salience and availability effects. We also observe the representativeness heuristic: atypical but valid attribute combinations are frequently flagged as erroneous, and in entity matching tasks, surface similarity produces a substantial false-positive rate with high confidence. In data repair, participants show a robust preference for leaving values missing rather than imputing plausible values, consistent with omission bias. In contrast, automation-aligned switching under strong contradiction does not exceed a conservative rare-error tolerance threshold at the population level, indicating that deference to automated recommendations is limited in this setting. Across scenarios, bias patterns persist among technically experienced participants and across diverse workflow practices, suggesting that bias in data cleaning reflects general cognitive tendencies rather than lack of expertise. These findings motivate human-in-the-loop cleaning systems that clearly separate representation from semantics, present expert or algorithmic recommendations non-prescriptively, and support reflective evaluation of atypical but valid cases.

cs.DB↗

arXiv2025

Relation-Stratified Sampling for Shapley Values Estimation in Relational Databases

Shapley-like values, including the Shapley and Banzhaf values, provide a principled way to quantify how individual tuples contribute to a query result. Their exact computation, however, is intractable because it requires aggregating marginal contributions over exponentially many permutations or subsets. While sampling-based estimators have been studied in cooperative game theory, their direct use for relational query answering remains underexplored and often ignores the structure of schemas and joins. We study tuple-level attribution for relational queries through sampling and introduce Relation-Stratified Sampling (RSS). Instead of stratifying coalitions only by size, RSS partitions the sample space by a relation-wise count vector that records how many tuples are drawn from each relation. This join-aware stratification concentrates samples on structurally valid and informative coalitions and avoids strata that cannot satisfy query conditions. We further develop an adaptive variant, ARSS, that reallocates budget across strata using variance estimates obtained during sampling, improving estimator efficiency without increasing the total number of samples. We analyze these estimators, describe a practical implementation that reuses compiled views to reduce per-sample query cost, and evaluate them on TPCH workloads. Across diverse queries with multi-relation joins and aggregates, RSS and ARSS consistently outperform classical Monte Carlo (MCS) and size-based Stratified Sampling (SS), yielding lower error and variance with fewer samples. An ablation shows that relation-aware stratification and adaptive allocation contribute complementary gains, making ARSS a simple, effective, and anytime estimator for database-centric Shapley attribution.

cs.DB↗

arXiv2025

Beyond Accuracy: An Empirical Study of Uncertainty Estimation in Imputation

Handling missing data is a central challenge in data-driven analysis. Modern imputation methods not only aim for accurate reconstruction but also differ in how they represent and quantify uncertainty. Yet, the reliability and calibration of these uncertainty estimates remain poorly understood. This paper presents a systematic empirical study of uncertainty in imputation, comparing representative methods from three major families: statistical (MICE, SoftImpute), distribution alignment (OT-Impute), and deep generative (GAIN, MIWAE, TabCSDI). Experiments span multiple datasets, missingness mechanisms (MCAR, MAR, MNAR), and missingness rates. Uncertainty is estimated through three complementary routes: multi-run variability, conditional sampling, and predictive-distribution modeling, and evaluated using calibration curves and the Expected Calibration Error (ECE). Results show that accuracy and calibration are often misaligned: models with high reconstruction accuracy do not necessarily yield reliable uncertainty. We analyze method-specific trade-offs among accuracy, calibration, and runtime, identify stable configurations, and offer guidelines for selecting uncertainty-aware imputers in data cleaning and downstream machine learning pipelines.

cs.DB↗

arXiv2025

Heterogeneity in Entity Matching: A Survey and Experimental Analysis

Entity matching (EM) is a fundamental task in data integration and analytics, essential for identifying records that refer to the same real-world entity across diverse sources. In practice, datasets often differ widely in structure, format, schema, and semantics, creating substantial challenges for EM. We refer to this setting as Heterogeneous EM (HEM). This survey offers a unified perspective on HEM by introducing a taxonomy, grounded in prior work, that distinguishes two primary categories -- representation and semantic heterogeneity -- and their subtypes. The taxonomy provides a systematic lens for understanding how variations in data form and meaning shape the complexity of matching tasks. We then connect this framework to the FAIR principles -- Findability, Accessibility, Interoperability, and Reusability -- demonstrating how they both reveal the challenges of HEM and suggest strategies for mitigating them. Building on this foundation, we critically review recent EM methods, examining their ability to address different heterogeneity types, and conduct targeted experiments on state-of-the-art models to evaluate their robustness and adaptability under semantic heterogeneity. Our analysis uncovers persistent limitations in current approaches and points to promising directions for future research, including multimodal matching, human-in-the-loop workflows, deeper integration with large language models and knowledge graphs, and fairness-aware evaluation in heterogeneous settings.

cs.DB↗

arXiv2025

Fuzzy Ontology Embeddings and Visual Query Building for Ontology Exploration

Ontologies play a central role in structuring knowledge across domains, supporting tasks such as reasoning, data integration, and semantic search. However, their large size and complexity, particularly in fields such as biomedicine, computational biology, law, and engineering, make them difficult for non-experts to navigate. Formal query languages such as SPARQL offer expressive access but require users to understand the ontology's structure and syntax. In contrast, visual exploration tools and basic keyword-based search interfaces are easier to use but often lack flexibility and expressiveness. We introduce FuzzyVis, a proof-of-concept system that enables intuitive and expressive exploration of complex ontologies. FuzzyVis integrates two key components: a fuzzy logic-based querying model built on fuzzy ontology embeddings, and an interactive visual interface for building and interpreting queries. Users can construct new composite concepts by selecting and combining existing ontology concepts using logical operators such as conjunction, disjunction, and negation. These composite concepts are matched against the ontology using fuzzy membership-based embeddings, which capture degrees of membership and support approximate, concept-level similarity search. The visual interface supports browsing, query composition, and partial search without requiring formal syntax. By combining fuzzy semantics with embedding-based reasoning, FuzzyVis enables flexible interpretation, efficient computation, and exploratory learning. Case studies demonstrate how FuzzyVis supports subtle information needs and helps users uncover relevant concepts in large, complex ontologies.

cs.HC↗

arXiv2025

Adaptive Anomaly Detection in the Presence of Concept Drift: Extended Report

The presence of concept drift poses challenges for anomaly detection in time series. While anomalies are caused by undesirable changes in the data, differentiating abnormal changes from varying normal behaviours is difficult due to differing frequencies of occurrence, varying time intervals when normal patterns occur, and identifying similarity thresholds to separate the boundary between normal vs. abnormal sequences. Differentiating between concept drift and anomalies is critical for accurate analysis as studies have shown that the compounding effects of error propagation in downstream tasks lead to lower detection accuracy and increased overhead due to unnecessary model updates. Unfortunately, existing work has largely explored anomaly detection and concept drift detection in isolation. We introduce AnDri, a framework for Anomaly detection in the presence of Drift. AnDri introduces the notion of a dynamic normal model where normal patterns are activated, deactivated or newly added, providing flexibility to adapt to concept drift and anomalies over time. We introduce a new clustering method, Adjacent Hierarchical Clustering (AHC), for learning normal patterns that respect their temporal locality; critical for detecting short-lived, but recurring patterns that are overlooked by existing methods. Our evaluation shows AnDri outperforms existing baselines using real datasets with varying types, proportions, and distributions of concept drift and anomalies.

cs.DB↗

arXiv2024

Reducing Biases in Record Matching Through Scores Calibration

Record matching models typically output a real-valued matching score that is later consumed through thresholding, ranking, or human review. While fairness in record matching has mostly been assessed using binary decisions at a fixed threshold, such evaluations can miss systematic disparities in the entire score distribution and can yield conclusions that change with the chosen threshold. We introduce a threshold-independent notion of score bias that extends standard group-fairness criteria-demographic parity (DP), equal opportunity (EO), and equalized odds (EOD)-from binary outputs to score functions by integrating group-wise metric gaps over all thresholds. Using this metric, we empirically show that several state-of-the-art deep matchers can exhibit substantial score bias even when appearing fair at commonly used thresholds. To mitigate these disparities without retraining the underlying matcher, we propose two model-agnostic post-processing methods that only require score evaluations on an (unlabeled) calibration set. Calib targets DP by aligning minority/majority score distributions to a common Wasserstein barycenter via a quantile-based optimal-transport map, with finite-sample guarantees on both residual DP bias and score distortion. C-Calib extends this idea to label-dependent notions (EO/EOD) by performing barycenter alignment conditionally on an estimated label, and we characterize how its guarantees depend on both sample size and label-estimation error. Experiments on standard record-matching benchmarks and multiple neural matchers confirm that Calib and C-Calib substantially reduce score bias with minimal loss in accuracy.

cs.LG↗

arXiv2024

Evaluating SQL Understanding in Large Language Models

The rise of large language models (LLMs) has significantly impacted various domains, including natural language processing (NLP) and image generation, by making complex computational tasks more accessible. While LLMs demonstrate impressive generative capabilities, there is an ongoing debate about their level of "understanding," particularly in structured domains like SQL. In this paper, we evaluate the extent to which LLMs "understand" SQL by testing them on a series of key SQL tasks. These tasks, such as syntax error detection, missing token identification, query performance prediction, query equivalence checking, and query explanation, assess the models' proficiency in recognition, context awareness, semantics, and coherence, which are essential skills for SQL understanding. We generate labeled datasets from well-known workloads, and evaluate the latest LLMs, focusing on how query complexity and syntactic features influence performance. Our results indicate that while GPT4 excels at tasks requiring recognition and context, all models struggle with deeper semantic understanding and coherence, especially in query equivalence and performance estimation, revealing the limitations of current LLMs in achieving full SQL comprehension.

cs.DB↗

arXiv2024

Evaluating Blocking Biases in Entity Matching

Entity Matching (EM) is crucial for identifying equivalent data entities across different sources, a task that becomes increasingly challenging with the growth and heterogeneity of data. Blocking techniques, which reduce the computational complexity of EM, play a vital role in making this process scalable. Despite advancements in blocking methods, the issue of fairness; where blocking may inadvertently favor certain demographic groups; has been largely overlooked. This study extends traditional blocking metrics to incorporate fairness, providing a framework for assessing bias in blocking techniques. Through experimental analysis, we evaluate the effectiveness and fairness of various blocking methods, offering insights into their potential biases. Our findings highlight the importance of considering fairness in EM, particularly in the blocking phase, to ensure equitable outcomes in data integration tasks.

cs.LG↗

arXiv2024

Threshold-Independent Fair Matching through Score Calibration

Entity Matching (EM) is a critical task in numerous fields, such as healthcare, finance, and public administration, as it identifies records that refer to the same entity within or across different databases. EM faces considerable challenges, particularly with false positives and negatives. These are typically addressed by generating matching scores and apply thresholds to balance false positives and negatives in various contexts. However, adjusting these thresholds can affect the fairness of the outcomes, a critical factor that remains largely overlooked in current fair EM research. The existing body of research on fair EM tends to concentrate on static thresholds, neglecting their critical impact on fairness. To address this, we introduce a new approach in EM using recent metrics for evaluating biases in score based binary classification, particularly through the lens of distributional parity. This approach enables the application of various bias metrics like equalized odds, equal opportunity, and demographic parity without depending on threshold settings. Our experiments with leading matching methods reveal potential biases, and by applying a calibration technique for EM scores using Wasserstein barycenters, we not only mitigate these biases but also preserve accuracy across real world datasets. This paper contributes to the field of fairness in data cleaning, especially within EM, which is a central task in data cleaning, by promoting a method for generating matching scores that reduce biases across different thresholds.

cs.LG↗

arXiv2024

OTClean: Data Cleaning for Conditional Independence Violations using Optimal Transport

Ensuring Conditional Independence (CI) constraints is pivotal for the development of fair and trustworthy machine learning models. In this paper, we introduce \sys, a framework that harnesses optimal transport theory for data repair under CI constraints. Optimal transport theory provides a rigorous framework for measuring the discrepancy between probability distributions, thereby ensuring control over data utility. We formulate the data repair problem concerning CIs as a Quadratically Constrained Linear Program (QCLP) and propose an alternating method for its solution. However, this approach faces scalability issues due to the computational cost associated with computing optimal transport distances, such as the Wasserstein distance. To overcome these scalability challenges, we reframe our problem as a regularized optimization problem, enabling us to develop an iterative algorithm inspired by Sinkhorn's matrix scaling algorithm, which efficiently addresses high-dimensional and large-scale data. Through extensive experiments, we demonstrate the efficacy and efficiency of our proposed methods, showcasing their practical utility in real-world data cleaning and preprocessing tasks. Furthermore, we provide comparisons with traditional approaches, highlighting the superiority of our techniques in terms of preserving data utility while ensuring adherence to the desired CI constraints.

cs.LG↗

arXiv2023

Semi-Oblivious Chase Termination for Linear Existential Rules: An Experimental Study

The chase procedure is a fundamental algorithmic tool in databases that allows us to reason with constraints, such as existential rules, with a plethora of applications. It takes as input a database and a set of constraints, and iteratively completes the database as dictated by the constraints. A key challenge, though, is the fact that it may not terminate, which leads to the problem of checking whether it terminates given a database and a set of constraints. In this work, we focus on the semi-oblivious version of the chase, which is well-suited for practical implementations, and linear existential rules, a central class of constraints with several applications. In this setting, there is a mature body of theoretical work that provides syntactic characterizations of when the chase terminates, algorithms for checking chase termination, precise complexity results, and worst-case optimal bounds on the size of the result of the chase (whenever is finite). Our main objective is to experimentally evaluate the existing chase termination algorithms with the aim of understanding which input parameters affect their performance, clarifying whether they can be used in practice, and revealing their performance limitations.

cs.DB↗

arXiv2021

Extending Sticky-Datalog+/- via Finite-Position Selection Functions: Tractability, Algorithms, and Optimization

Weakly-Sticky(WS) Datalog+/- is an expressive member of the family of Datalog+/- program classes that is defined on the basis of the conditions of stickiness and weak-acyclicity. Conjunctive query answering (QA) over the WS programs has been investigated, and its tractability in data complexity has been established. However, the design and implementation of practical QA algorithms and their optimizations have been open. In order to fill this gap, we first study Sticky and WS programs from the point of view of the behavior of the chase procedure. We extend the stickiness property of the chase to that of generalized stickiness of the chase (GSCh) modulo an oracle that selects (and provides) the predicate positions where finitely values appear during the chase. Stickiness modulo a selection function S that provides only a subset of those positions defines sch(S), a semantic subclass of GSCh. Program classes with selection functions include Sticky and WS, and another syntactic class that we introduce and characterize, namely JWS, of jointly-weakly-sticky programs, which contains WS. The selection functions for these last three classes are computable, and no external, possibly non-computable oracle is needed. We propose a bottom-up QA algorithm for programs in the class sch(S), for a general selection function S. As a particular case, we obtain a polynomial-time QA algorithm for JWS and weakly-sticky programs. Unlike WS, JWS turns out to be closed under magic-sets query optimization. As a consequence, both the generic polynomial-time QA algorithm and its magic-set optimization can be particularized and applied to WS.

cs.DB↗

arXiv2020

Privacy-Aware Data Cleaning-as-a-Service (Extended Version)

Data cleaning is a pervasive problem for organizations as they try to reap value from their data. Recent advances in networking and cloud computing technology have fueled a new computing paradigm called Database-as-a-Service, where data management tasks are outsourced to large service providers. In this paper, we consider a Data Cleaning-as-a-Service model that allows a client to interact with a data cleaning provider who hosts curated, and sensitive data. We present PACAS: a Privacy-Aware data Cleaning-As-a-Service model that facilitates interaction between the parties with client query requests for data, and a service provider using a data pricing scheme that computes prices according to data sensitivity. We propose new extensions to the model to define generalized data repairs that obfuscate sensitive data to allow data sharing between the client and service provider. We present a new semantic distance measure to quantify the utility of such repairs, and we re-define the notion of consistency in the presence of generalized values. The PACAS model uses (X,Y,L)-anonymity that extends existing data publishing techniques to consider the semantics in the data while protecting sensitive values. Our evaluation over real data show that PACAS safeguards semantically related sensitive values, and provides lower repair errors compared to existing privacy-aware cleaning techniques.

cs.DB↗

arXiv2020

Diversifying Anonymized Data with Diversity Constraints

Recently introduced privacy legislation has aimed to restrict and control the amount of personal data published by companies and shared to third parties. Much of this real data is not only sensitive requiring anonymization, but also contains characteristic details from a variety of individuals. This diversity is desirable in many applications ranging from Web search to drug and product development. Unfortunately, data anonymization techniques have largely ignored diversity in its published result. This inadvertently propagates underlying bias in subsequent data analysis. We study the problem of finding a diverse anonymized data instance where diversity is measured via a set of diversity constraints. We formalize diversity constraints and study their foundations such as implication and satisfiability. We show that determining the existence of a diverse, anonymized instance can be done in PTIME, and we present a clustering-based algorithm. We conduct extensive experiments using real and synthetic data showing the effectiveness of our techniques, and improvement over existing baselines. Our work aligns with recent trends towards responsible data science by coupling diversity with privacy-preserving data publishing.

cs.DB↗

arXiv2020

Facilitating SQL Query Composition and Analysis

Formulating efficient SQL queries requires several cycles of tuning and execution, particularly for inexperienced users. We examine methods that can accelerate and improve this interaction by providing insights about SQL queries prior to execution. We achieve this by predicting properties such as the query answer size, its run-time, and error class. Unlike existing approaches, our approach does not rely on any statistics from the database instance or query execution plans. This is particularly important in settings with limited access to the database instance. Our approach is based on using data-driven machine learning techniques that rely on large query workloads to model SQL queries and their properties. We evaluate the utility of neural network models and traditional machine learning models. We use two real-world query workloads: the Sloan Digital Sky Survey (SDSS) and the SQLShare query workload. Empirical results show that the neural network models are more accurate in predicting the query error class, achieving a higher F-measure on classes with fewer samples as well as performing better on other problems such as run-time and answer size prediction. These results are encouraging and confirm that SQL query workloads and data-driven machine learning methods can be leveraged to facilitate query composition and analysis.

cs.DB↗