Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reducing Large Language Model Bias with Emphasis on 'Restricted Industries': Automated Dataset Augmentation and Prejudice Quantification

Devam Mondal, Carlo Lipizzi

TL;DR

A novel, automated mechanism for debiasing through specified dataset augmentation in the lens of bias producers and in the context of 'restricted industries' with limited data is proposed.

Abstract

Despite the growing capabilities of large language models, there exists concerns about the biases they develop. In this paper, we propose a novel, automated mechanism for debiasing through specified dataset augmentation in the lens of bias producers and in the context of 'restricted industries' with limited data. We additionally create two new additional metrics, the mb-index and db-index, to quantify bias, considering the idea that bias occurs due to both intrinsic model architecture and dataset.

Reducing Large Language Model Bias with Emphasis on 'Restricted Industries': Automated Dataset Augmentation and Prejudice Quantification

TL;DR

A novel, automated mechanism for debiasing through specified dataset augmentation in the lens of bias producers and in the context of 'restricted industries' with limited data is proposed.

Abstract

Despite the growing capabilities of large language models, there exists concerns about the biases they develop. In this paper, we propose a novel, automated mechanism for debiasing through specified dataset augmentation in the lens of bias producers and in the context of 'restricted industries' with limited data. We additionally create two new additional metrics, the mb-index and db-index, to quantify bias, considering the idea that bias occurs due to both intrinsic model architecture and dataset.
Paper Structure (15 sections, 6 equations, 2 figures, 2 tables)

This paper contains 15 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Cluster 1: Types and Examples of Bias in the Realm of LLMs
  4. Cluster 2: Bias in Application of LLMs in Restricted Industries
  5. Cluster 3: Dataset Bias in the Realm of LLMs
  6. Cluster 4: Inherent Bias in LLM Architectures
  7. Addressing and Remediating Bias
  8. Approach
  9. Dataset Augmentation
  10. LLM Bias Classification
  11. Dataset Bias Classification
  12. Method
  13. Results and Discussion
  14. Limitations and Future Research Directions
  15. Conclusion

Figures (2)

  • Figure 1: Obtaining semantically homogenous clusters through k-means clustering and hyperparameter grid search.
  • Figure 2: Obtaining semantically homogenous clusters through k-means clustering and hyperparameter grid search.