ADEPT: A DEbiasing PrompT Framework

Authors: Ke Yang, Charles Yu, Yi R. Fung, Manling Li, Heng Ji

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate ADEPT on several widely acknowledged debiasing benchmarks and downstream tasks, and find that it achieves competitive results while maintaining (and in some cases even improving) the PLM s representation ability.
Researcher Affiliation Academia Ke Yang1, Charles Yu2, Yi R. Fung2, Manling Li2, Heng Ji2 1Tsinghua University 2University of Illinois Urbana-Champaign
Pseudocode Yes Algorithm 1: ADEPT: a debiasing algorithm for contextualized word embeddings.
Open Source Code Yes 1The code and data are publicly available at https://github.com/ Empath Yang/ADEPT.
Open Datasets Yes For the sentences associated with the word tuples, we draw sentences from News-Commentary v15 (Tiedemann 2012) for the gender setting and sentences from Book Corpus (Zhu et al. 2015) and News-Commentary v15 (Tiedemann 2012) for the relgions setting. Since the original Book Corpus is no longer available, we use (lewtun et al. 2022) which is an open source replica.
Dataset Splits No The paper explicitly mentions test examples for evaluation but does not provide specific details on a validation dataset split used during the training or hyperparameter tuning of ADEPT itself.
Hardware Specification Yes All the experiments are conducted on two Ge Force RTX 3090 GPUs and in a Linux operating system.
Software Dependencies No The paper mentions using BERT-LARGE-UNCASED from Hugging Face and the Adam optimizer but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We set λ in Equation 1 to be 7 and ρ in Equation 4 to be 15. We use Adam (Kingma and Ba 2014) to optimize the objective function. During the debiasing process, our learning rate is 5e-5 and our batchsize is 32.