A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Filters
"Thy algorithm shalt not bear false witness": An Evaluation of Multiclass Debiasing Methods on Word Embeddings
[article]
2020
arXiv
pre-print
Preserved Fulltext
Other Versions
Specifically, in the natural language processing domain, it has been shown that social biases persist in word embeddings and are thus in danger of amplifying these biases when used. ...
As an example of social bias, religious biases are shown to persist in word embeddings and the need for its removal is highlighted. ...
In general the results above show that the word embedding ConceptNet carries the least bias as evaluated by MAC and RNSB scores. ...
arXiv:2010.16228v2
fatcat:l5gd43dua5cbxgg5akoahcp6m4
Improving QA Generalization by Concurrent Modeling of Multiple Biases
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Existing NLP datasets contain various biases that models can easily exploit to achieve high performances on the corresponding evaluation sets. ...
In this paper, we investigate the impact of debiasing methods for improving generalization and propose a general framework for improving the performance on both in-domain and out-of-domain datasets by ...
The first author of the paper is supported by a scholarship from Technical University of Darmstadt. ...
arXiv:2010.03338v1
fatcat:b5xbxfiixfewpo5x26r5suanf4
Dictionary-based Debiasing of Pre-trained Word Embeddings
[article]
2021
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
Unlike prior work, our proposed method does not require the types of biases to be pre-defined in the form of word lists, and learns the constraints that must be satisfied by unbiased word embeddings automatically ...
of the word according to the dictionary, and (c) remains orthogonal to the vector space spanned by any biased basis vectors in the pre-trained word embedding space. ...
Words in evaluation sets T3, T4 and T8 are not covered by the input pre-trained embeddings and hence not considered in this evaluation. ...
arXiv:2101.09525v1
fatcat:uqepsduj4zfebmngdqloefymyu
Towards Debiasing NLU Models from Unknown Biases
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Other Versions
In this work, we present the first step to bridge this gap by introducing a self-debiasing framework that prevents models from mainly utilizing biases without knowing them in advance. ...
Recently proposed debiasing methods are shown to be effective in mitigating this tendency. ...
This work is funded by the German Research Foundation through the research training group AIPHES (GRK 1994/1) and by the German Federal Ministry of Education and Research and the Hessen State Ministry ...
arXiv:2009.12303v4
fatcat:to6ealbpvjftdc2ksrfzp6xqau
Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings
2019
Proceedings of the 2019 Conference of the North
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Next, we propose a novel methodology for the evaluation of multiclass debiasing. We demonstrate that our multiclass debiasing is robust and maintains the efficacy in standard NLP tasks. ...
Word embeddings, trained on these texts, perpetuate and amplify these stereotypes, and propagate biases to machine learning models that use word embeddings as features. ...
Finally, we are greatly appreciative of the anonymous reviewers for their time and constructive comments. ...
doi:10.18653/v1/n19-1062
dblp:conf/naacl/ManziniLBT19
fatcat:2fv5rpdwovggxjebk4r2gsv5um
An Empirical Survey of the Effectiveness of Debiasing Techniques for Pre-trained Language Models
[article]
2022
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
Note that this fulltext copy is not of the "primary" version of this work. The version it corresponds to is:
2021 pre-print arXiv:2110.08527v1 fatcat:2bzih3q5pzgq3a24yxpmke753eOther Versions
non-gender biases; And (3) improvements on bias benchmarks such as StereoSet and CrowS-Pairs by using debiasing strategies are often accompanied by a decrease in language modeling ability, making it difficult ...
This has attracted attention to developing techniques that mitigate such biases. ...
Acknowledgements SR is supported by the Canada CIFAR AI Chairs program and the NSERC Discovery Grant program. NM is supported by an IVADO Excellence Scholarship. ...
arXiv:2110.08527v3
fatcat:o6rk7ui6jvhfzlbphfvifgkjem
Learning from Failure: Training Debiased Classifier from Biased Classifier
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Other Versions
that go against the prejudice of the biased network in (a). ...
Based on the observations, we propose a failure-based debiasing scheme by training a pair of neural networks simultaneously. ...
In parallel, we train a "debiased" neural network by focusing on samples that the biased model struggles to learn, which are expected to be samples conflicting with the bias. ...
arXiv:2007.02561v2
fatcat:d5i6mgy2pzatrixenen4avx54u
Black is to Criminal as Caucasian is to Police: Detecting and Removing Multiclass Bias in Word Embeddings
[article]
2019
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Note that this fulltext copy is not of the "primary" version of this work. The version it corresponds to is:
2019 pre-print arXiv:1904.04047v2 fatcat:7jqikpoymnftvigvmf3eawggteOther Versions
Next, we propose a novel methodology for the evaluation of multiclass debiasing. We demonstrate that our multiclass debiasing is robust and maintains the efficacy in standard NLP tasks. ...
Word embeddings, trained on these texts, perpetuate and amplify these stereotypes, and propagate biases to machine learning models that use word embeddings as features. ...
Finally, we are greatly appreciative of the anonymous reviewers for their time and constructive comments. ...
arXiv:1904.04047v3
fatcat:xhhisxxfxjfzxoyza3bes3krlm
Fair Embedding Engine: A Library for Analyzing and Mitigating Gender Bias in Word Embeddings
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Further, it will allow rapid prototyping of new methods by evaluating their performance on a suite of standard metrics. ...
FEE will aid practitioners in fast track analysis of existing debiasing methods on their embedding models. ...
Since all debiasing methods are fundamentally applied to a word embedding, the class of each debiasing method is initialised by a WE object. ...
arXiv:2010.13168v1
fatcat:owlv4r2cxvh57fb2y2uldppaku
MDR Cluster-Debias: A Nonlinear WordEmbedding Debiasing Pipeline
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Existing methods for debiasing word embeddings often do so only superficially, in that words that are stereotypically associated with, e.g., a particular gender in the original embedding space can still ...
This indicates that word embeddings encode gender bias in still other ways, not necessarily captured by upstream tests. ...
For bias-based evaluation, we use the same six cluster-based bias measures that are proposed by Gonen and Goldberg as our upstream bias-based evaluation tasks. The first one we call Kmeans Accuracy. ...
arXiv:2006.11642v1
fatcat:6mfebvhfuranzbg74atemyjkne
End-to-End Bias Mitigation by Modelling Biases in Corpora
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Other Versions
During training, the bias-only models' predictions are used to adjust the loss of the base model to reduce its reliance on biases by down-weighting the biased examples and focusing the training on the ...
Results show that our debiasing methods greatly improve robustness in all settings and better transfer to other textual entailment datasets. Our code and data are publicly available in . ...
Let x b i ∈ X b be biased features of x i that are predictive of y i . ...
arXiv:1909.06321v3
fatcat:mlewy4vggvervkdjl6in3ibcwm
A General Framework for Implicit and Explicit Debiasing of Distributional Word Vector Spaces
2020
PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
Moreover, we successfully transfer debiasing models, by means of cross-lingual embedding spaces, and remove or attenuate biases in distributional word vector spaces of languages that lack readily available ...
We then propose three debiasing models that operate on explicit or implicit bias specifications and that can be composed towards more robust debiasing. ...
14 We show only the input space and the spaces debiased with GBDD and BAM. We provide similar illustrations for other debiasing models in the supplementary material. ...
doi:10.1609/aaai.v34i05.6325
fatcat:4w5b52ls3bhurbfjrpgajcg3c4
A General Framework for Implicit and Explicit Debiasing of Distributional Word Vector Spaces
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
Other Versions
Moreover, we successfully transfer debiasing models, by means of cross-lingual embedding spaces, and remove or attenuate biases in distributional word vector spaces of languages that lack readily available ...
We then propose three debiasing models that operate on explicit or implicit bias specifications and that can be composed towards more robust debiasing. ...
14 We show only the input space and the spaces debiased with GBDD and BAM. We provide similar illustrations for other debiasing models in the supplementary material. ...
arXiv:1909.06092v2
fatcat:vtrcm6iemjazznflhzoardacba
Sustainable Modular Debiasing of Language Models
[article]
2021
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
We showcase ADELE, in gender debiasing of BERT: our extensive evaluation, encompassing three intrinsic and two extrinsic bias measures, renders ADELE, very effective in bias mitigation. ...
To remedy for this, a wide range of debiasing techniques have recently been introduced to remove such stereotypical biases from PLMs. ...
In the following, we focus on approaches for mitigating biases from PLMs, which are largely inspired by debiasing for static word embeddings (e.g., Bolukbasi et al., 2016; Dev and Phillips, 2019; Lauscher ...
arXiv:2109.03646v1
fatcat:cpjfhauztnatnddh4dohzj5ogi
RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Models
[article]
2021
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
Our results indicate that DialoGPT is biased with respect to religious groups and that some debiasing techniques can remove this bias while preserving downstream task performance. ...
Text representation models are prone to exhibit a range of societal biases, reflecting the non-controlled and biased nature of the underlying pretraining data, which consequently leads to severe ethical ...
Our experimental results indicate that (i) DialoGPT is significantly biased along two (out of five) bias evaluation dimensions and (ii) that some of the employed debiasing methods (see §4) manage to reduce ...
arXiv:2106.03521v1
fatcat:s43dffsi25goppqnb742qxcf7e
« Previous
Showing results 1 — 15 out of 4,360 results