Saved in:
| Main Authors: | , |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2025
|
| Online Access: | https://doi.org/10.5281/zenodo.18106999 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <p>At present, anomaly and bias detection have emerged as two of the most crucial problems in state-of-the-art Artificial Intelligence (AI) /Machine Learning (ML) systems. This has become more important today because there are already many new AI-based systems being used in certain fields of life, such as: health care, finance, cyber security, autonomous vehicles and smart infrastructure. Both anomaly detection and bias detection constitute areas of growing importance in AI systems deployed to high-stakes domains, where reliability, safety, and fairness are crucial. Both anomalies and biases can signal deeper problems, a data quality issue, change in distribution, or inequitable model behavior but despite such commonalities their methods of origin have similar statistical background, methods for representation learning and even procedures of model distribution. In this paper we present a com-prehensive overview of anomaly detection and bias detection methodologies and techniques employed in AI systems. As this survey takes deep dives into both anomaly detection and bias detection, it gives an extensive view on these closely related fields in parallel. A combination of conventional statistical approaches with modern Deep Learning (DL) insights, along with hybrid models that collect aspects of physics, and different framework for designing, analyzing and operating AI systems the paper presents a unified perspective that clarifies the relationships, methodologies, and challenges across these domains. It also highlights the gaps of past research, identifies new avenues for innovation, and sets out best practices to make sure AI is reliable, fair, and transparent. In doing so this survey will be useful to policy-makers, researchers or people who are working in practical positions and want the transparency of AI's near-term future.</p>