A Survey on Bias in Visual Datasets

Fabbrizzi, Simone; Papadopoulos, Symeon; Ntoutsi, Eirini; Kompatsiaris, Ioannis

doi:10.48550/arxiv.2107.07919

Cited by 2 publications

(2 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The problem of finding the "right" data: For acquiring data, data scientists have to rely on data mining with little to no quality checking and potential biases [4]. Biased datasets are a common cause for erroneous or unexpected behaviour of ML models in critical environments, such as in medical diagnostic [8], in the juridical system [19,37], or in safety-critical applications [15,46]. There are attempts to create "unbiased" datasets.…”

Section: Related Literaturementioning

confidence: 99%

An investigation of challenges encountered when specifying training data and runtime monitors for safety critical ML applications

Heyn¹,

Knauss²,

Malleswaran³

et al. 2023

Preprint

View full text Add to dashboard Cite

The development and operation of critical software that contains machine learning (ML) models requires diligence and established processes. Especially the training data used during the development of ML models have major influences on the later behaviour of the system. Runtime monitors are used to provide guarantees for that behaviour. [Question / problem] We see major uncertainty in how to specify training data and runtime monitoring for critical ML models and by this specifying the final functionality of the system. In this interview-based study we investigate the underlying challenges for these difficulties. [Principal ideas/results] Based on ten interviews with practitioners who develop ML models for critical applications in the automotive and telecommunication sector, we identified 17 underlying challenges in 6 challenge groups that relate to the challenge of specifying training data and runtime monitoring.[Contribution] The article provides a list of the identified underlying challenges related to the difficulties practitioners experience when specifying training data and runtime monitoring for ML models. Furthermore, interconnection between the challenges were found and based on these connections recommendation proposed to overcome the root causes for the challenges.

show abstract

Section: Related Literaturementioning

confidence: 99%

An investigation of challenges encountered when specifying training data and runtime monitors for safety critical ML applications

Heyn¹,

Knauss²,

Malleswaran³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Fabbrizi et al [6] present an overview of the major biases encountered in computer vision tasks that include selection, framing and label biases. Selection bias can be characterised as "any disparities or associations created as a result of the process by which subjects are included in a visual dataset" [6]. Selection bias is encountered in numerous ML models.…”

Section: Bias In Computer Vision Algorithmsmentioning

confidence: 99%

Bias in Face Image Classification Machine Learning Models: The Impact of Annotator’s Gender and Race

Kafkalias

Herodotou

Θεοδοσίου

et al. 2022

IFIP Advances in Information and Communication Technology

View full text Add to dashboard Cite

An important factor that ensures the correct operation of Machine Learning models is the quality of data used during the model training process. Quite often, training data is annotated by humans, and as a result, annotation bias may be introduced. In this study, we focus on face image classification and aim to quantify the effect of annotation bias introduced by different groups of annotators, allowing in that way the understanding of the problems that arise due to annotation bias. The results of the experiments indicate that the performance of Machine Learning models in several face image interpretation tasks is correlated to the self-reported demographic characteristics of the annotators. In particular, we found significant correlation to annotator race, while correlation to gender is less profound. Furthermore, experimental results show that it is possible to determine the group of annotators involved in the annotation process by considering the annotation data provided by previously unseen annotators. The results emphasize the risks of annotation bias in Machine Learning models.

show abstract

A Survey on Bias in Visual Datasets

Cited by 2 publications

References 61 publications

An investigation of challenges encountered when specifying training data and runtime monitors for safety critical ML applications

An investigation of challenges encountered when specifying training data and runtime monitors for safety critical ML applications

Bias in Face Image Classification Machine Learning Models: The Impact of Annotator’s Gender and Race

Contact Info

Product

Resources

About