AI language models have gotten a necessary a part of our lives. Now we have been using Google for a long time to access information, but now, we’re slowly switching to ChatGPT. It provides concise answers, clear explanations, and it is generally quicker to seek out the knowledge we seek.
These models learn from the information we produced over time. Consequently, we transferred our biases to the AI models, and this can be a topic of debate within the domain. One particular bias that has gained attention is the gender bias in pronoun distributions, where models are likely to prefer gendered pronouns corresponding to “he” or “she” based on the context.
Addressing this gender bias is crucial for ensuring fair and inclusive language generation. For instance, should you start the sentence “The CEO believes that…”, the model continues with he, and should you replace the CEO with the nurse, the following token becomes she. This instance serves as an interesting case study to look at biases and explore methods to mitigate them.
It seems that the context plays an important role in shaping these biases. By replacing CEO with a occupation stereotypically related to a special gender, we will actually flip the observed bias. But here’s the challenge: achieving consistent debiasing across all different contexts where CEO appears isn’t any easy task. We would like interventions that work reliably and predictably, no matter the particular situation. In any case, interpretability and control are key in terms of understanding and improving language models. Unfortunately, the present Transformer models, while impressive of their performance, don’t quite meet these criteria. Their contextual representations introduce all types of complex and nonlinear effects that depend upon the context at hand.
So, how can we overcome these challenges? How can we tackle the bias we introduced in large language models? Should we improve transformers, or should we give you latest structures? The reply is Backpack Language Models.
Backpack LM tackles the challenge of debiasing pronoun distributions by leveraging non-contextual representations often known as sense vectors. These vectors capture different facets of a word’s meaning and its role in diverse contexts, giving words multiple personalities.
In Backpack LMs, predictions are log-linear mixtures of non-contextual representations, known as sense vectors. Each word within the vocabulary is represented by multiple sense vectors, encoding distinct learned facets of the word’s potential roles in numerous contexts.
These sense vectors specialize and might be predictively useful in specific contexts. The weighted sum of sense vectors for words in a sequence forms the Backpack representation of every word, with the weights determined by a contextualization function that operates on all the sequence. By leveraging these sense vectors, Backpack models enable precise interventions that behave predictably across all contexts.
Which means that we will make non-contextual changes to the model that consistently influences its behavior. In comparison with Transformer models, Backpack models offer a more transparent and manageable interface. They supply precise interventions which are easier to grasp and control. Furthermore, Backpack models don’t compromise on performance either. Actually, they achieve results on par with Transformers while offering enhanced interpretability.
Sense vectors in Backpack models encode wealthy notions of word meaning, outperforming word embeddings of state-of-the-art Transformer models on lexical similarity tasks. Moreover, interventions on sense vectors, corresponding to reducing gender bias in skilled words, show the control mechanism offered by Backpack models. By downscaling the sense vector related to gender bias, significant reductions in contextual prediction disparities might be achieved in limited settings.
Check Out The Paper and Project. Don’t forget to hitch our 24k+ ML SubReddit, Discord Channel, and Email Newsletter, where we share the most recent AI research news, cool AI projects, and more. If you’ve any questions regarding the above article or if we missed anything, be at liberty to email us at Asif@marktechpost.com
Featured Tools From AI Tools Club
🚀 Check Out 100’s AI Tools in AI Tools Club
Ekrem Çetinkaya received his B.Sc. in 2018, and M.Sc. in 2019 from Ozyegin University, Istanbul, Türkiye. He wrote his M.Sc. thesis about image denoising using deep convolutional networks. He received his Ph.D. degree in 2023 from the University of Klagenfurt, Austria, together with his dissertation titled “Video Coding Enhancements for HTTP Adaptive Streaming Using Machine Learning.” His research interests include deep learning, computer vision, video encoding, and multimedia networking.