Fixing Gender Bias in Google Translate (Bias in Neural Machine Translation)
That's right, Google Translate is biased. When it comes to assigning gender from a gender-neutral language like Hungarian, it defaults to the stereotypical translations. But why does it do this? And how can we fix this?
We're diving into a LOT of interesting Bias in Neural Machine Translation research, from how we evaluate dataset bias to the solution I think is the most promising way of debiasing our large language models!
Transcript on my blog: https://mukulrathi.com/google-translate-is-biased/
------------------------------------------------
Social:
Twitter: https://twitter.com/mukulrathi_
My website (+ blog): https://mukulrathi.com
My email newsletter: https://newsletter.mukulrathi.com
------------------------------------------------
Links:
What constitutes data bias?: https://www.youtube.com/watch?v=MgZFSny8GjI&ab_channel=MukulRathi
Google Translate Bias: https://arxiv.org/abs/1809.02208
Google AI blog post: https://ai.googleblog.com/2020/04/a-scalable-approach-to-reducing-gender.html
Intellectual Debt / ML Systems Design by Neil Lawrence: http://inverseprobability.com/talks/notes/deploying-machine-learning-systems-intellectual-debt-and-auto-ai.html
WinoMT paper: https://www.aclweb.org/anthology/P19-1164.pdf
Getting Gender Right in Neural Machine Translation: https://arxiv.org/abs/1909.05088
Domain Adaptation for Gender Debiasing: https://arxiv.org/abs/2004.04498
Neural Machine Translation Doesn’t Translate Gender Coreference Right
Unless You Make It: https://arxiv.org/pdf/2010.05332.pdf
------------------------------------------------------------------------------------
Timestamps:
00:00 Introduction
00:30 Overview
00:59 What causes bias?
03:09 Pipelines aren't the solution
05:33 Evaluating Bias (WinoMT)
08:04 Fixing the data
09:56 Domain Adaptation for Gender Debiasing
----------------------------------------
► Artist Attribution
Music By: "After The Fall"
Track Name: "Silk"
Published by: Chill Out Records
- Source: https://goo.gl/fh3rEJ
Official After The Fall YouTube Channel Below
https://www.youtube.com/channel/UCGQE...
License: Eric/After The Fall @ Just Chill Productions
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)
Full license here: https://creativecommons.org/licenses
► Download "Silk" by 'After The Fall' HERE: https://www.chilloutrecordsmusic.com
Видео Fixing Gender Bias in Google Translate (Bias in Neural Machine Translation) канала Mukul Rathi
We're diving into a LOT of interesting Bias in Neural Machine Translation research, from how we evaluate dataset bias to the solution I think is the most promising way of debiasing our large language models!
Transcript on my blog: https://mukulrathi.com/google-translate-is-biased/
------------------------------------------------
Social:
Twitter: https://twitter.com/mukulrathi_
My website (+ blog): https://mukulrathi.com
My email newsletter: https://newsletter.mukulrathi.com
------------------------------------------------
Links:
What constitutes data bias?: https://www.youtube.com/watch?v=MgZFSny8GjI&ab_channel=MukulRathi
Google Translate Bias: https://arxiv.org/abs/1809.02208
Google AI blog post: https://ai.googleblog.com/2020/04/a-scalable-approach-to-reducing-gender.html
Intellectual Debt / ML Systems Design by Neil Lawrence: http://inverseprobability.com/talks/notes/deploying-machine-learning-systems-intellectual-debt-and-auto-ai.html
WinoMT paper: https://www.aclweb.org/anthology/P19-1164.pdf
Getting Gender Right in Neural Machine Translation: https://arxiv.org/abs/1909.05088
Domain Adaptation for Gender Debiasing: https://arxiv.org/abs/2004.04498
Neural Machine Translation Doesn’t Translate Gender Coreference Right
Unless You Make It: https://arxiv.org/pdf/2010.05332.pdf
------------------------------------------------------------------------------------
Timestamps:
00:00 Introduction
00:30 Overview
00:59 What causes bias?
03:09 Pipelines aren't the solution
05:33 Evaluating Bias (WinoMT)
08:04 Fixing the data
09:56 Domain Adaptation for Gender Debiasing
----------------------------------------
► Artist Attribution
Music By: "After The Fall"
Track Name: "Silk"
Published by: Chill Out Records
- Source: https://goo.gl/fh3rEJ
Official After The Fall YouTube Channel Below
https://www.youtube.com/channel/UCGQE...
License: Eric/After The Fall @ Just Chill Productions
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)
Full license here: https://creativecommons.org/licenses
► Download "Silk" by 'After The Fall' HERE: https://www.chilloutrecordsmusic.com
Видео Fixing Gender Bias in Google Translate (Bias in Neural Machine Translation) канала Mukul Rathi
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![Counterfactuals - the solution to fairness and explainability?](https://i.ytimg.com/vi/G4kWljXZCHY/default.jpg)
![Advances and Open Problems in Federated Learning | Paper Explained](https://i.ytimg.com/vi/0ttnWfj-OLo/default.jpg)
![Confidential and Private Collaborative Learning (CaPC)](https://i.ytimg.com/vi/u5lTu7W7RJM/default.jpg)
![Privacy-preserving ML with PATE](https://i.ytimg.com/vi/jm1Sfdno_5A/default.jpg)
![Intro to Explainable AI](https://i.ytimg.com/vi/2W7DwqJ9b-M/default.jpg)
![Attacks on Neural Networks EXPLAINED](https://i.ytimg.com/vi/wyHY2U6IsWw/default.jpg)
![Deep Learning with Differential Privacy (DP-SGD explained)](https://i.ytimg.com/vi/oNSelFJnPaM/default.jpg)
![Federated Learning EXPLAINED (Tutorial + Research + Frameworks)](https://i.ytimg.com/vi/nBGQQHPkyNY/default.jpg)
![Why I Chose Industry over a PhD](https://i.ytimg.com/vi/CehDs7KQuH4/default.jpg)
![Does debiasing word embeddings actually work? (+ explanation of GN-GloVe, Hard-debiasing)](https://i.ytimg.com/vi/MKs2io7opJs/default.jpg)
![What constitutes dataset bias? (and what can we do about it?)](https://i.ytimg.com/vi/MgZFSny8GjI/default.jpg)