Software fairness

Brun, Yuriy; Meliou, Alexandra

doi:10.1145/3236024.3264838

Cited by 86 publications

(42 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Evaluating Models' Bias/Fairness. Evaluating the bias and fairness of a system is important both from a theoretical and a practical perspective [7,41,88,89]. Related studies first define a fairness criteria and then try to optimize the original objective while satisfying the fairness criteria [3,12,13,21,36,46].…”

Section: Related Workmentioning

confidence: 99%

Testing DNN image classifiers for confusion & bias errors

Tian

Zhong

Ordóñez

et al. 2020

Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering

View full text Add to dashboard Cite

Image classifiers are an important component of today's software, from consumer and business applications to safety-critical domains. The advent of Deep Neural Networks (DNNs) is the key catalyst behind such widespread success. However, wide adoption comes with serious concerns about the robustness of software systems dependent on DNNs for image classification, as several severe erroneous behaviors have been reported under sensitive and critical circumstances. We argue that developers need to rigorously test their software's image classifiers and delay deployment until acceptable. We present an approach to testing image classifier robustness based on class property violations. We found that many of the reported erroneous cases in popular DNN image classifiers occur because the trained models confuse one class with another or show biases towards some classes over others. These bugs usually violate some class properties of one or more of those classes. Most DNN testing techniques focus on perimage violations, so fail to detect class-level confusions or biases. We developed a testing technique to automatically detect classbased confusion and bias errors in DNN-driven image classification software. We evaluated our implementation, DeepInspect, on several popular image classifiers with precision up to 100% (avg. 72.6%) for confusion errors, and up to 84.3% (avg. 66.8%) for bias errors. DeepInspect found hundreds of classification mistakes in widelyused models, many exposing errors indicating confusion or bias. CCS CONCEPTS • Software and its engineering → Software testing and debugging; • Computing methodologies → Neural networks.

show abstract

Section: Related Workmentioning

confidence: 99%

Testing DNN image classifiers for confusion & bias errors

Tian

Zhong

Ordóñez

et al. 2020

Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering

View full text Add to dashboard Cite

show abstract

“…In addition, Brun et al insists on the important point in the information system development in consideration of fairness as follows. "(1) eliciting and specifying requirements that capture fairness properties, (2) architecting and designing systems while adhering to their fairness concerns, (3) validating and verifying fairness properties of the resulting software products, and (4) maintaining the fairness properties as the software evolves" [5]．Our proposed method contributes to these solutions.…”

Section: B Fairnessmentioning

confidence: 99%

“…However, a small number of opinions may be ignored, as in many cases they will generally make a good choice. In recent years, fairness has become important in the world [5], and it is better to be able to select goals in consideration of this concept.…”

Section: Introductionmentioning

confidence: 99%

Fairness Analysis in Goal-Oriented Requirements Selection

Horita¹,

Noguchi²

2020

IJTEF

View full text Add to dashboard Cite

Requirements analysis is an important step in business information system development. Goal-oriented requirements analysis is a powerful method that captures requirements to be satisfied in business information systems as goals and derives means for achieving goals or alternative means. The difficulty in this analysis is that it is difficult to analyze manually when the combination of goals is huge, and to take into account the preferences of the various stakeholders. It is difficult to satisfy the requirements of all stakeholders at the same time because preferences are different and trade-offs for each stakeholder. Therefore, there is a risk that the requirements of some stakeholders may be ignored. Therefore, in this paper, we propose a method using multi-objective optimization to take into account stakeholder preference and to make goal selection with emphasis on fairness. Our method performs a goal selection having both high utility and fairness of the entire stakeholders. We evaluated by a case study and showed the effectiveness.

show abstract

“…o f so ftw are en g in ee rin g fo r m ach in e learn in g [2], [3], such as te s t g en eratio n fo r d etectin g fairn ess v io latio n s [5 ]- [7 ], data has greater fairness. Indeed, we found that in 28% of the cases, a larger set of training data even has greater unfairness than a smaller set.…”

mentioning

confidence: 99%

"Ignorance and Prejudice" in Software Fairness

Zhang

Harman

2021

2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE)

View full text Add to dashboard Cite

Machine learning software can be unfair when making human-related decisions, having prejudices over certain groups of people. Existing work primarily focuses on proposing fairness metrics and presenting fairness improvement approaches. It remains unclear how key aspect of any machine learning system, such as feature set and training data, affect fairness. This paper presents results from a comprehensive study that addresses this problem. We find that enlarging the feature set plays a significant role in fairness (with an average effect rate of 38%). Importantly, and contrary to widely-held beliefs that greater fairness often corresponds to lower accuracy, our findings reveal that an enlarged feature set has both higher accuracy and fairness. Perhaps also surprisingly, we find that a larger training data does not help to improve fairness. Our results suggest a larger training data set has more unfairness than a smaller one when feature sets are insufficient; an important cautionary finding for practising software engineers.

show abstract

Software fairness

Cited by 86 publications

References 64 publications

Testing DNN image classifiers for confusion & bias errors

Testing DNN image classifiers for confusion & bias errors

Fairness Analysis in Goal-Oriented Requirements Selection

"Ignorance and Prejudice" in Software Fairness

Contact Info

Product

Resources

About