Search for: All records

The plethora of fairness metrics developed for ranking-based decision-making raises the question: which metrics align best with people’s perceptions of fairness, and why? Most prior studies examining people’s perceptions of fairness metrics tend to use ordinal rating scales (e.g., Likert scales). However, such scales can be ambiguous in their interpretation across participants, and can be influenced by interface features used to capture responses.We address this gap by exploring the use of two-alternative forced choice methodologies— used extensively outside the fairness community for comparing visual stimuli— to quantitatively compare participant perceptions across fairness metrics and ranking characteristics. We report a crowdsourced experiment with 224 participants across four conditions: two alternative rank fairness metrics, ARP and NDKL, and two ranking characteristics, lists of 20 and 100 candidates, resulting in over 170,000 individual judgments. Quantitative results show systematic differences in how people interpert these metrics, and surprising exceptions where fairness metrics disagree with people’s perceptions. Qualitative analyses of participant comments reveals an interplay between cognitive and visual strategies that affects people’s perceptions of fairness. From these results, we discuss future work in aligning fairness metrics with people’s perceptions, and highlight the need and benefits of expanding methodologies for fairness studies. 

Rated preference aggregation is conventionally performed by averaging ratings from multiple evaluators to create a consensus ordering of candidates from highest to lowest average rating. Ideally, the consensus is fair, meaning critical opportunities are not withheld from marginalized groups of candidates, even if group biases may be present in the to-be-combined ratings. Prior work operationalizing fairness in preference aggregation is limited to settings where evaluators provide rankings of candidates (e.g., Joe > Jack > Jill). Yet, in practice, many evaluators assign ratings such as Likert scales or categories (e.g., yes, no, maybe) to each candidate. Ratings convey different information than rankings leading to distinct fairness issues during their aggregation. The existing literature does not characterize these fairness concerns nor provide applicable bias-mitigation solutions. Unlike the ranked setting studied previously, two unique forms of bias arise in rating aggregation. First, biased rating stems from group disparities in to-be-aggregated evaluator ratings. Second, biased tie-breaking occurs because ties in average ratings must be resolved when aggregating ratings into a consensus ranking, and this tie-breaking act can unfairly advantage certain groups. To address this gap, we define the open fair rated preference aggregation problem and introduce the corresponding Fate methodology. Fate offers the first group fairness metric specifically for rated preference data. We propose two Fate algorithms. Fate-Break works in settings when ties need to be broken, explicitly fairness-enhancing such processes without lowering consensus utility. Fate-Rate mitigates disparities in how groups are rated, by using a Markov-chain approach to generate outcomes where groups are, in as much as possible, equally represented. Our experimental study illustrates the FATE methods provide the most bias-mitigation compared to adapting prior methods to fair tie-breaking and rating aggregation. 

We present FairRankTune, a multi-purpose open-source Python toolkit offering three primary services: quantifying fairness-related harms, leveraging bias mitigation algorithms, and constructing custom fairness-relevant datasets. FairRankTune provides researchers and practitioners with a self-contained resource for fairness auditing, experimentation, and advancing research. The central piece of FairRankTune is a novel fairness-tunable ranked data generator, RankTune, that streamlines the creation of custom fairness-relevant ranked datasets. FairRankTune also offers numerous fair ranking metrics and fairness-aware ranking algorithms within the same plug-and-play package. We demonstrate the key innovations of FairRankTune, focusing on features that are valuable to stakeholders via use cases highlighting workflows in the end-to-end process of mitigating bias in ranking systems. FairRankTune addresses the gap of limited publicly available datasets, auditing tools, and implementations for fair ranking. 

Due to the scarcity of reliable anomaly labels, recent anomaly detection methods leveraging noisy auto-generated labels either select clean samples or refurbish noisy labels. However, both approaches struggle due to the unique properties of anomalies.Sample selectionoften fails to separate sufficiently many clean anomaly samples from noisy ones, whilelabel refurbishmenterroneously refurbishesmarginalclean samples. To overcome these limitations, we design Unity, thefirstlearning from noisy labels (LNL) approach for anomaly detection that elegantly leverages the merits of both sample selection and label refurbishment to iteratively prepare a diverse clean sample set for network training. Unity uses a pair of deep anomaly networks to collaboratively select samples with clean labels based on prediction agreement, followed by a disagreement resolution mechanism to capture marginal samples with clean labels. Thereafter, Unity utilizes unique properties of anomalies to design an anomaly-centric contrastive learning strategy that accurately refurbishes the remaining noisy labels. The resulting set, composed ofselected and refurbishedclean samples, will be used to train the anomaly networks in the next training round. Our experimental study on 10 real-world benchmark datasets demonstrates that Unity consistently outperforms state-of-the-art LNL techniques by up to 0.31 in F-1 Score (0.52 \rightarrow 0.83). 

As learning-to-rank models are increasingly deployed for decision-making in areas with profound life implications, the FairML community has been developing fair learning-to-rank (LTR) models. These models rely on the availability of sensitive demographic features such as race or sex. However, in practice, regulatory obstacles and privacy concerns protect this data from collection and use. As a result, practitioners may either need to promote fairness despite the absence of these features or turn to demographic inference tools to attempt to infer them. Given that these tools are fallible, this paper aims to further understand how errors in demographic inference impact the fairness performance of popular fair LTR strategies. In which cases would it be better to keep such demographic attributes hidden from models versus infer them? We examine a spectrum of fair LTR strategies ranging from fair LTR with and without demographic features hidden versus inferred to fairness-unaware LTR followed by fair re-ranking. We conduct a controlled empirical investigation modeling different levels of inference errors by systematically perturbing the inferred sensitive attribute. We also perform three case studies with real-world datasets and popular open-source inference methods. Our findings reveal that as inference noise grows, LTR-based methods that incorporate fairness considerations into the learning process may increase bias. In contrast, fair re-ranking strategies are more robust to inference errors. All source code, data, and experimental artifacts of our experimental study are available here: https://github.com/sewen007/hoiltr.git 

Poster. 

Preference aggregation mechanisms help decision-makers combine diverse preference rankings produced by multiple voters into a single consensus ranking. Prior work has developed methods for aggregating multiple rankings into a fair consensus over the same set of candidates. Yet few real-world problems present themselves as such precisely formulated aggregation tasks with each voter fully ranking all candidates. Instead, preferences are often expressed as rankings over partial and even disjoint subsets of candidates. For instance, hiring committee members typically opt to rank their top choices instead of exhaustively ordering every single job applicant. However, the existing literature does not offer a framework for characterizing nor ensuring group fairness in such partial preference aggregation tasks. Unlike fully ranked settings, partial preferences imply both a selection decision of whom to rank plus an ordering decision of how to rank the selected candidates. Our work fills this gap by conceptualizing the open problem of fair partial preference aggregation. We introduce an impossibility result for fair selection from partial preferences and design a computational framework showing how we can navigate this obstacle. Inspired by Single Transferable Voting, our proposed solution PreFair produces consensus rankings that are fair in the selection of candidates and also in their relative ordering. Our experimental study demonstrates that PreFair achieves the best performance in this dual fairness objective compared to state-of-the-art alternatives adapted to this new problem while still satisfying voter preferences. 

Log anomaly detection, critical in identifying system failures and preempting security breaches, finds irregular patterns within large volumes of log data. Modern log anomaly detectors rely on training deep learning models on clean anomaly-free log data. However, such clean log data requires expensive and tedious human labeling. In this paper, we thus propose a robust log anomaly detection framework, PlutoNOSPACE, that automatically selects a clean representative sample subset of the polluted log sequence data to train a Transformer-based anomaly detection model. Pluto features three innovations. First, due to localized concentrations of anomalies inherent in the embedding space of log data, Pluto partitions the sequence embedding space generated by the model into regions that then allow it to identify and discard regions that are highly polluted by our pollution level estimation scheme, based on our pollution quantification via Gaussian mixture modeling. Second, for the remaining more slightly polluted regions, we select samples that maximally purify the eigenvector spectrum, which can be transformed into the NP-hard facility location problem; allowing us to leverage its greedy solution with a (1-(1/e)) approximation guarantee in optimality. Third, by iteratively alternating between the above subset selection, a model re-training on the latest subset, and a subset filtering using dynamic training artifacts generated by the latest model, the data selected is progressively refined. The final sample set is used to retrain the final anomaly detection model. Our experiments on four real-world log benchmark datasets demonstrate that by retaining 77.7% (BGL) to 96.6% (ThunderBird) of the normal sequences while effectively removing 90.3% (BGL) to 100.0% (ThunderBird, HDFS) of the anomalies, Pluto provides a significant absolute F-1 improvement up to 68.86% (2.16% → 71.02%) compared to the state-of-the-art sample selection methods. The implementation of this work is available at https://github.com/LeiMa0324/Pluto-SIGMOD25. 

Log anomaly detection, critical in identifying system failures and preempting security breaches, finds irregular patterns within large volumes of log data. Modern log anomaly detectors rely on training deep learning models on clean anomaly-free log data. However, such clean log data requires expensive and tedious human labeling. In this paper, we thus propose a robust log anomaly detection framework, PlutoNOSPACE, that automatically selects a clean representative sample subset of the polluted log sequence data to train a Transformer-based anomaly detection model. Pluto features three innovations. First, due to localized concentrations of anomalies inherent in the embedding space of log data, Pluto partitions the sequence embedding space generated by the model into regions that then allow it to identify and discard regions that are highly polluted by our pollution level estimation scheme, based on our pollution quantification via Gaussian mixture modeling. Second, for the remaining more slightly polluted regions, we select samples that maximally purify the eigenvector spectrum, which can be transformed into the NP-hard facility location problem; allowing us to leverage its greedy solution with a (1-(1/e)) approximation guarantee in optimality. Third, by iteratively alternating between the above subset selection, a model re-training on the latest subset, and a subset filtering using dynamic training artifacts generated by the latest model, the data selected is progressively refined. The final sample set is used to retrain the final anomaly detection model. Our experiments on four real-world log benchmark datasets demonstrate that by retaining 77.7\% (BGL) to 96.6\% (ThunderBird) of the normal sequences while effectively removing 90.3\% (BGL) to 100.0\% (ThunderBird, HDFS) of the anomalies, Pluto provides a significant absolute F-1 improvement up to 68.86\% (2.16\% → 71.02\%) compared to the state-of-the-art sample selection methods. The implementation of this work is available at https://github.com/LeiMa0324/Pluto-SIGMOD25.