# Fighting biases with dynamic boosting

**Authors:** Anna Veronika Dorogush, Andrey Gulin, Gleb Gusev, Nikita Kazeev, Liudmila Ostroumova Prokhorenkova, Aleksandr Vorobev  
**arXiv:** [1706.09516](https://arxiv.org/abs/1706.09516)  
**Date:** 2017-06-28  
**Summary page:** https://www.summarizepaper.com/en/paper/1706.09516v1/

## Key points

- Gradient boosting algorithms in machine learning are prone to label leakage, leading to systematic bias in pointwise gradient estimates.
- The paper "Fighting biases with dynamic boosting" by Anna Veronika Dorogush et al. addresses this issue and proposes solutions to mitigate the bias.
- The authors introduce a dynamic boosting algorithm that outperforms existing implementations and enhances model accuracy.
- Experimental results show that the proposed approach not only addresses bias but also improves model effectiveness, offering a valuable advancement for practitioners and researchers in the field of machine learning.

## Summary

In their paper titled "Fighting biases with dynamic boosting," authors Anna Veronika Dorogush, Andrey Gulin, Gleb Gusev, Nikita Kazeev, Liudmila Ostroumova Prokhorenkova, and Aleksandr Vorobev address a critical issue in gradient boosting algorithms. Gradient boosting is widely used in modern industrial machine learning and data science applications; however, existing implementations are prone to label leakage that introduces systematic bias in pointwise gradient estimates. This ultimately leads to reduced accuracy in the generated models. The authors conduct a formal analysis of this issue and propose solutions to mitigate the bias by incorporating a dynamic boosting algorithm into the open-source implementation of gradient boosting. Experimental results demonstrate that this new approach not only addresses the bias problem but also outperforms popular implementations currently available. Overall, "Fighting biases with dynamic boosting" offers a significant contribution to the field of machine learning by highlighting and providing solutions for a crucial limitation in current gradient boosting algorithms. The proposed approach has the potential to enhance model accuracy and effectiveness using gradient boosting techniques, making it a valuable advancement for practitioners and researchers in the field.