Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Towards Foundation Auto-Encoders for Time-Series Anomaly Detection

Created by
  • Haebom

Author

Gast on Garc ia Gonz alez, Pedro Casas, Emilio Mart inez, Alicia Fern andez

Outline

In this paper, we present a novel time series modeling approach, building on the success of large-scale pre-training-based models. We introduce Foundation Auto-Encoders (FAEs), a generative AI model for anomaly detection in time series data based on variational autoencoders (VAEs). The term “foundation” refers to a model that has been pre-trained on a large amount of time series data to learn complex temporal patterns and thereby perform accurate modeling, prediction, and anomaly detection on previously unseen datasets. FAEs utilize VAEs and dilated convolutional neural networks (DCNNs) to build a general model for univariate time series modeling, which ultimately can perform appropriately in out-of-the-box zero-shot anomaly detection applications. In this paper, we introduce the main concepts of FAEs and present preliminary results on several multidimensional time series datasets from various domains, including a real-world dataset from an operational mobile ISP and the well-known KDD 2021 anomaly detection dataset.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel approach to apply large-scale pre-training-based models to time series anomaly detection.
◦
Building a general model for univariate time series modeling by combining VAEs and DCNNs.
◦
Presenting possibilities for zero-shot anomaly detection applications.
◦
Preliminary results are presented on real-world datasets from various domains.
•
Limitations:
◦
Only preliminary results are presented, requiring further validation of the model's performance and generalization ability.
◦
The results of application to multidimensional time series data are limited and further research is needed.
◦
Lack of analysis of model complexity and computational cost.
◦
Performance evaluation and stability verification in actual application environments are required.
👍
Made with Slashpage