search

ruiking04 / CutAddPaste

CutAddPaste: Time Series Anomaly Detection by Exploiting Abnormal Knowledge
3 stars 1 forks source link

High similarity between CutAddPaste and AnomalyBERT #2

Open LFM-bot opened 1 day ago

LFM-bot commented 1 day ago

The basic idea and methodology of CutAddPaste are highly similar to the ICLR23 paper AnomalyBERT [1]. However, there is no reference to AnomalyBERT and I can't find any discussion regarding the differences between these two methods. It is Interesting.

[1] Jeong Y, Yang E, Ryu J H, et al. Anomalybert: Self-supervised transformer for time series anomaly detection using data degradation scheme[J]. arXiv preprint arXiv:2305.04468, 2023.

ruiking04 commented 1 day ago

I read the above paper and I have the following 4 points to clarify:

  1. The two papers are different in both the abnormal augmentation methods and network structures
  2. Our article is mainly inspired by CutPaste[2]. In addition, the above article does not cite CutPaste[2].
  3. Our article was submitted to AAAI-2024 in 2023 for the first time, see the figure below.
  4. Our subsequent paper will discuss the paper mentioned in the issue.

[2] Li, Chun-Liang, et al. "Cutpaste: Self-supervised learning for anomaly detection and localization." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2021.

image

LFM-bot commented 1 day ago

Thanks for the prompt reply. I'm looking forward to your follow-up work.