Related Experiment Video
Updated: May 23, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
8.9K
Evaluating the method reproducibility of deep learning models in biodiversity research
Waqas Ahmed1, Vamsi Krishna Kommineni1,2,3, Birgitta König-Ries1,2,4
1Heinz Nixdorf Chair of Distributed Information Systems, Friedrich-Schiller Universität Jena, Jena, Thuringia, Germany.
Peerj. Computer Science
|March 10, 2025
Summary
Reproducibility in artificial intelligence (AI) for biodiversity research is vital. This study found that while datasets are shared in half of publications, many lack crucial details on deep learning methods and randomness.
Area of Science:
- Ecology
- Computer Science
- Conservation Biology
Background:
- Artificial intelligence (AI) is transforming biodiversity research through advanced data analysis and monitoring.
- Reproducibility is essential for the credibility and verification of AI-driven ecological findings.
Purpose of the Study:
- To investigate the reproducibility of deep learning (DL) methods in biodiversity research.
- To develop and apply a methodology for evaluating the reproducibility of DL techniques in ecological publications.
Main Methods:
- Defined ten variables for method reproducibility across four categories: resource requirements, methodological information, uncontrolled randomness, and statistical considerations.
- Manually extracted reproducibility variable availability from 100 biodiversity publications using DL.
- Developed a tiered system for classifying reproducibility levels based on variable availability.
Main Results:
- Datasets were shared in 50% of the analyzed publications.
- A significant portion of publications lacked comprehensive details on deep learning methods, particularly concerning randomness.
- Identified key areas where information is missing for ensuring reproducibility.
Conclusions:
- While AI offers powerful tools for biodiversity research, current practices often fall short in providing sufficient information for full reproducibility.
- Improvements in reporting methodological details, especially regarding randomness, are needed to enhance the reliability of AI applications in ecology.
- Standardized reporting guidelines could bolster the credibility and impact of AI-driven biodiversity conservation efforts.
