Paper introduces metrics to detect unintended bias in text classifiers.
problem Unintended bias in machine learning classifiers.
method Threshold-agnostic metrics considering various score distribution variations across groups.
result New metrics reveal subtle unintended bias in public models.
The Pinned AUC metric hides unintended bias when class distributions vary.
problem Unintended bias in classification models.
method Examines the Pinned AUC metric and its limitations.
result Pinned AUC can obscure different types of unintended bias.
This work introduces a framework to detect unintended bias in facial analysis models.
problem Detecting unintended biases in facial analysis models used in critical applications.
method Image counterfactual sensitivity analysis using generative adversarial networks.
result Identifies factors affecting facial classifier predictions, revealing unintended biases.
Mitigates bias in text classification by weighting instances.
problem Unintended biases in text classification datasets based on demographic terms.
method Instance weighting to recover non-discrimination distribution.
result Effective mitigation of unintended biases without sacrificing generalization.
Study shows FL reduces unintended memorization by clustering data and using strong user-level privacy.
problem Unintended memorization in federated learning.
method Examined the effect of clustering data and using strong user-level differential privacy in FL.
result Clustering data and strong user-level differential privacy reduce unintended memorization.
Unintended effects from scaling neural network outputs with adaptive learning rates.
problem Adaptive learning rate optimization's behavior is altered by output scaling, leading to misinterpretation.
method Presented a modified optimization algorithm to mitigate unintended effects.
result Adaptive learning rate's effectiveness is significantly impacted by output scaling, especially for small scaling factors.
Paper bridges AI/ML and causal modeling to reduce bias.
problem Difficulty in combining methods from different assumptions.
method Integrates system dynamics and structural equation modeling.
result Unified mathematical framework for AI/ML and causal modeling.
Proposes a test to ensure predictive algorithms predict intended outcomes better than unintended ones.
problem Unintended model behavior leading to prediction of unintended outcomes.
method Falsification framework using nonparametric hypothesis testing to compare prediction losses across outcomes.
result Establishes discriminant validity with respect to gender but not race in an admissions setting.
This work identifies and mitigates reasoning shortcuts in Neuro-Symbolic models.
problem Neuro-Symbolic models can achieve high accuracy by using unintended concepts.
method Characterized reasoning shortcuts as unintended optima of the learning objective and identified four key conditions.
result Reasoning shortcuts are difficult to mitigate, casting doubt on NeSy solutions' trustworthiness and interpretability.
GRASP removes spurious correlations in fine-tuned models, improving task performance and reducing bias.
problem Fine-tuned models can latch onto spurious correlations, leading to bias and reduced generalization.
method GRASP identifies and removes spurious correlations from model weights without removing latent factors.
result GRASP significantly reduces bias and improves task performance in various fine-tuning tasks.
SPAT improves adversarial robustness by preserving semantics in adversarial training.
problem Adversarial examples often have different semantics than original data, introducing unintended biases.
method Semantics-preserving adversarial training (SPAT) that encourages pixel perturbation shared among all classes.
result SPAT improves adversarial robustness and achieves state-of-the-art results in CIFAR-10 and CIFAR-100.
Fairness in ML models can lead to counterintuitive predictions.
problem Enforcing fairness in machine learning models leads to unexpected outcomes.
method Introducing slack-consistency as a desirable property for fairness procedures.
result Standard fairness methods violate the property of monotonicity with respect to slack.
Improved scalability and interpretability in training data attribution.
problem Identifying which training data drives specific behaviors, especially unintended ones.
method Leveraging interpretable structures within the model to attribute model behavior to semantic directions, not individual test examples.
result Simple probe-based attribution methods are first-order approximations of Concept Influence that achieve comparable performance while being over an order-of-magnitude faster.
We look at the effect of the tick size changes on the TOPIX 100 index names made by the Tokyo Stock Exchange on Jan-14-2014 and Jul-22-2104. The intended consequence of the change is price improvement and shorter time to execution. We look at security level metrics that include the spread, trading volume, number of tra…
The paper proposes trading grades in a financial market to address unintended consequences of grading systems.
problem Unintended consequences of grading systems, such as unfair advantages and misaligned incentives.
method A thought experiment in a financial market structure to trade grades, similar to interest rate swaps.
result Grades should be viewed as personal equity, not used for selection criteria.
New findings reveal discount regularization can be seen as a strong prior, leading to poor performance in unevenly sampled data.
problem Discount regularization leads to poor performance in unevenly sampled data.
method Equivalence theorem showing discount regularization as a strong prior, setting regularization parameters locally for individual state-action pairs.
result Discount regularization can be seen as a strong prior, leading to poor performance in unevenly sampled data.
Proposes a falsification framework to test algorithmic discriminant validity.
problem Unintended model behavior in predictive algorithms.
method Falsification framework based on statistical tests comparing prediction losses across outcomes.
result Establishes discriminant validity for some outcomes but not others.
Fairness is a critical trait in decision making. As machine-learning models are increasingly being used in sensitive application domains (e.g. education and employment) for decision making, it is crucial that the decisions computed by such models are free of unintended bias. But how can we automatically validate the fa…
Symbolic knowledge in neural models can inadvertently make them more vulnerable to adversarial attacks.
problem Symbolic knowledge in neural models can make models more susceptible to adversarial attacks.
method Investigated deep probabilistic graphical models that incorporate symbolic knowledge and neural nets.
result Symbolic knowledge can propagate the negative effects of adversarial examples, making models more vulnerable.
Machine unlearning can compromise privacy, study shows.
problem Machine unlearning may leave data imprints in ML models, risking privacy.
method Proposed a membership inference attack to detect leakage.
result Machine unlearning can lead to unintended privacy risks.
Decisions taken in our everyday lives are based on a wide variety of information so it is generally very difficult to assess what are the strategies that guide us. Stock market therefore provides a rich environment to study how people take decision since responding to market uncertainty needs a constant update of these…
BC-Aligner maintains backward compatibility of embeddings after frequent updates.
problem Updating embeddings without requiring consumer teams to retrain their models.
method Learning backward compatible embeddings through BC-Aligner.
result BC-Aligner maintains backward compatibility with existing unintended tasks after multiple model version updates.
New attacks exploit transfer learning to misclassify text models.
problem Misclassification attacks against transfer learned text classifiers.
method Novel attack algorithms using unintended features from teacher models.
result Transfer learning increases vulnerability to misclassification attacks.
Paper introduces MinDiff framework for balancing classifier performance and fairness.
problem Balancing classifier performance and fairness in machine learning models.
method MinDiff framework with kernel-based statistical dependency tests.
result Demonstrates real-world improvements in classifier performance and fairness.
The paper proposes a test for trading intelligence similar to the Turing Test.
problem Determining if trading programs exhibit intelligent behavior.
method Develops a test based on the Turing Test for financial trading programs.
result Introduces a method to assess trading intelligence in financial markets.
New method stops experiments early for harm in diverse groups.
problem Early stopping of experiments for harmful treatment effects in diverse populations.
method Causal machine learning approach (CLASH) for early stopping.
result CLASH effectively stops experiments early for harmful treatment effects in diverse groups.
Previous studies have found that an adversary attacker can often infer unintended input information from intermediate-layer features. We study the possibility of preventing such adversarial inference, yet without too much accuracy degradation. We propose a generic method to revise the neural network to boost the challe…
New framework quantifies and reduces concept-based models' leakage.
problem Information leakage in concept-based models reduces interpretability.
method Information-theoretic framework with CTL and ICL measures.
result Measures predict model behaviour and identify leakage causes.
Study shows online learning algorithms incentivize low-quality content, proposing new algorithms to improve quality.
problem Online learning algorithms in content recommender systems incentivize producers to create low-quality content.
method Analyzed the game between producers and content quality, designed new learning algorithms to incentivize high effort and quality.
result New algorithms incentivize producers to invest high effort and achieve high user welfare, improving content quality.
The paper tests semantic importance in opaque models using betting.
problem Precise statistical guarantees for semantic concepts in black-box models.
method Formalizes global and local statistical importance via conditional independence and SKIT.
result Shows effectiveness and flexibility of the framework on various models.
New method uses coupled SDEs to edit images with high fidelity and consistency.
problem Challenges in editing image content with text-to-image models.
method Using coupled stochastic differential equations to guide generative model sampling.
result Achieves high prompt fidelity and near-pixel-level consistency.
This article guides data scientists on avoiding discrimination in machine learning.
problem Machine learning systems can create or exacerbate societal disparities.
method Provides a taxonomy of practices and measures to mitigate discrimination.
result Data scientists should be intentional about modeling and reducing discriminatory outcomes.
The study proposes an audit to assess user control over recommendations in collaborative filtering systems.
problem The gap between maximizing accuracy and ensuring user control over information availability in recommender systems.
method The approach involves a computationally efficient audit for top-N linear recommender models, focusing on reachability and user agency. result The study demonstrates that model complexity affects the effort required for users to exert control over their recommendations.
Research shows bias in machine learning can be due to algorithmic flaws, not just data.
problem Underestimation bias in machine learning algorithms.
method Initial research to understand factors contributing to bias in classification algorithms.
result Regularization methods to address overfitting can also accentuate bias.
Paper improves DOA estimation in sparse arrays using Siamese neural networks.
problem Challenges in DOA estimation with limited snapshots in sparse linear arrays.
method Introduces a Siamese neural network with a sparse augmentation layer for enhanced signal feature embedding.
result Demonstrates improved DOA estimation accuracy in sparse arrays.
Study proposes framework for selective classification using uncertainty quantification.
problem Unintended consequences of deep learning in selective classification.
method Mixed-integer programming framework combining model uncertainty and predictive mean.
result Framework outperforms industry standard methods significantly for online fraud management.
Financial LLMs need explicit bias consideration to avoid invalid results.
problem Finance-specific biases inflate performance and contaminate backtests.
method Identified five recurring biases and proposed a Structural Validity Framework.
result Explicit bias consideration is necessary for valid deployment claims.
We study the phenomenon of bias amplification in classifiers, wherein a machine learning model learns to predict classes with a greater disparity than the underlying ground truth. We demonstrate that bias amplification can arise via an inductive bias in gradient descent methods that results in the overestimation of the…
It has been noticed that some external CVIs exhibit a preferential bias towards a larger or smaller number of clusters which is monotonic (directly or inversely) in the number of clusters in candidate partitions. This type of bias is caused by the functional form of the CVI model. For example, the popular Rand index (R…
Mitigates gender bias amplification in model predictions.
problem Gender bias amplification in model predictions.
method Posterior regularization to mitigate bias.
result Almost removes gender bias amplification in model predictions.
Paper addresses reward learning issues in RL, improving both under- and over-estimation.
problem Reward learning from data can lead to reward delusions or underestimation, causing unintended behaviors.
method Connects reward learning to positive-unlabeled (PU) learning and applies a large-scale PU learning algorithm.
result Improves both GAIL and supervised reward learning without additional assumptions.
Depth uncertainty networks don't improve with bias correction, contrary to expectations.
problem Improving performance in active learning with overparameterised models like NNs.
method Depth uncertainty networks, compared to underparameterised models, show no improvement in performance with bias correction.
result Depth uncertainty networks do not improve with bias correction, unlike underparameterised models.
We quantify causal bias in continuous treatment settings.
problem Identifying and quantifying causal bias in continuous treatment scenarios.
method Developed a novel characterization of causal bias in structural causal models, proving conditions for zero bias and efficient estimation.
result Causal bias can be estimated efficiently under certain structural equation restrictions, allowing for causal regularization of predictive models.
The paper introduces Relative Bias to quantify LLM bias systematically.
problem Quantifying bias in LLMs is challenging due to ambiguity and rapid model emergence.
method Relative Bias framework using Embedding Transformation and LLM-as-a-Judge methodologies.
result The two scoring methods show strong alignment, providing a systematic approach.
Interpolated-MLPs control inductive bias for better performance in low-compute tasks.
problem Low-compute performance gap between MLPs and CNNs.
method Introduced Interpolated MLP (I-MLP) approach to control inductive bias incrementally.
result Continuous logarithmic relationship between inductive bias and performance in low-compute tasks.
SSMs have a built-in bias towards low-frequency components, which can be adjusted.
problem Frequency bias in SSMs affects their performance on long-range sequences.
method Proposed two mechanisms to tune frequency bias: scaling initialization or applying a Sobolev-norm-based filter.
result Tuning frequency bias improves SSMs' performance on long-range sequence learning tasks.
This paper assesses biases in contextualized word representations.
problem Analyzing biases in contextualized word representations.
method Proposes assessing bias at the contextual word level, capturing contextual effects of bias.
result Demonstrates evidence of bias in contextual word models, including racial bias and exacerbated effects for intersectional minorities.
A bias classifier is introduced to resist adversarial attacks.
problem Resisting adversarial attacks on deep neural networks (DNNs).
method Introducing the bias part of a DNN with Relu as the activation function as a classifier, and adding a random first-degree part to make it information-theoretically safe.
result The bias classifier is more robust than DNNs of similar size against adversarial attacks.