Debiasing Pre-Trained Language Models via Efficient Fine-Tuning
By Michael Gira, Ruisu Zhang, Kangwook Lee. 2022
49 citations recorded by the source corpus.
An explosion in the popularity of transformer-based language models (such as GPT-3, BERT, RoBERTa, and ALBERT) has opened the doors to new machine learning applications involving language modeling, text generation, and more. However, recent scrutiny reveals that these language models contain inherent biases towards certain demographics reflected in their training data. While research has tried mitigating this problem, existing approaches either fail to remove the bias completely, degrade performance (“catastrophic forgetting”), or are costly to execute. This work examines how to reduce gender bias in a GPT-2 language model by fine-tuning less than 1% of its parameters. Through quantitative benchmarks, we show that this is a viable way to reduce prejudice in pre-trained language models while remaining cost-effective at scale.
Subfields: Artificial Intelligence, Large Language Models, Safety, Robustness & Interpretability
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Hailey Schoelkopf
- Lintang Sutawika
- USVSN Sai Prashanth
- Shivanshu Purohit
- Quentin Anthony
- Stella Biderman
- Tong Yu
- Ryan A. Rossi
In Large Language Models
- Hailey Schoelkopf
- Lintang Sutawika
- USVSN Sai Prashanth
- Shivanshu Purohit
- Quentin Anthony
- Stella Biderman
- Neeraja Kirtane
- V. Manushree
In Safety, Robustness & Interpretability
- Stella Biderman
- Oskar van der Wal
- Neeraja Kirtane
- V. Manushree
- Aditya Kane
- Franck Dernoncourt
- Edward Raff
Citations in the corpus
- Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling (2023, methodology): “However, while some work has explored finetuning’s effects on bias in language models (Gira et al., 2022; Kirtane et al., 2022; Choenni et al., 2021), or the relationship between the corpus statistics and the measured bias (Bordia & Bowman, 2019; Van der Wal et al.”
- Bias and Fairness in Large Language Models: A Survey (2023, background): “Gira et al. (2022) freeze over 99% of a model’s parameters before fine-tuning on the WinoBias (Zhao et al., 2019) and CrowS-Pairs (Nangia et al., 2020) datasets, only updating a selective set of parameters, such as layer norm parameters or word positioning embeddings. Ranaldi et al. (2023) only upda”
- Efficient Gender Debiasing of Pre-trained Indic Language Models (2022, methodology): “Gira et al. (2022) fine-tuned a GPT model by only unfreezing some parameters to reduce the bias present.”