Typology of risks of generative text-to-image models
This paper investigates the direct risks and harms associated with modern text-to-image
generative models, such as DALL-E and Midjourney, through a comprehensive literature …
generative models, such as DALL-E and Midjourney, through a comprehensive literature …
The edinburgh international accents of english corpus: Towards the democratization of english asr
English is the most widely spoken language in the world, used daily by millions of people as
a first or second language in many different contexts. As a result, there are many varieties of …
a first or second language in many different contexts. As a result, there are many varieties of …
Pre-trained Speech Processing Models Contain Human-Like Biases that Propagate to Speech Emotion Recognition
I Slaughter, C Greenberg, R Schwartz… - arXiv preprint arXiv …, 2023 - arxiv.org
Previous work has established that a person's demographics and speech style affect how
well speech processing models perform for them. But where does this bias come from? In …
well speech processing models perform for them. But where does this bias come from? In …
Too brittle to touch: comparing the stability of quantization and distillation towards developing low-resource MT models
Leveraging shared learning through Massively Multilingual Models, state-of-the-art Machine
translation (MT) models are often able to adapt to the paucity of data for low-resource …
translation (MT) models are often able to adapt to the paucity of data for low-resource …
Just Because We Camp, Doesn't Mean We Should: The Ethics of Modelling Queer Voices
A Sigurgeirsson, EL Ungless - arXiv preprint arXiv:2406.07504, 2024 - arxiv.org
Modern voice cloning models claim to be able to capture a diverse range of voices. We test
the ability of a typical pipeline to capture the style known colloquially as" gay voice" and …
the ability of a typical pipeline to capture the style known colloquially as" gay voice" and …
Too Brittle To Touch: Comparing the Stability of Quantization and Distillation Towards Developing Lightweight Low-Resource MT Models
Leveraging shared learning through Massively Multilingual Models, state-of-the-art machine
translation models are often able to adapt to the paucity of data for low-resource languages …
translation models are often able to adapt to the paucity of data for low-resource languages …
Recalibrating machine learning for social biases: demonstrating a new methodology through a case study classifying gender biases in archival documentation
LJ Havens - 2024 - era.ed.ac.uk
This thesis proposes a recalibration of Machine Learning for social biases to minimize
harms from existing approaches and practices in the field. Prioritizing quality over quantity …
harms from existing approaches and practices in the field. Prioritizing quality over quantity …