Anomaly or simple misclassification in anomaly detection notebook

Gemini API
,
1

Hello,
I was going through some notebook examples, and I found the anomaly detection tile a bit misleading.
Basically, given some classes, what it does is to extract the embedding, run tsne, find the centroids for each class in kmeans fashions, pick a distance and everything over the distance is an outlier or anomaly.
I think anomaly detection from the literature is a different task, and anomaly is usually a sample OOD not ID.

3

Hi @Marco_Piccirilli

Welcome to forum! Can you share the notebook example that you are looking into?

Thanks

4

From the detailed description, I am pretty sure the OP is referring to Google Colab

Hope that helps

1 Like