VAE-based anomaly detection — advantages.
hardAnswer
- VAE learns probabilistic latent space with p(x).
- Anomaly score: (1) reconstruction error, (2) negative ELBO, (3) KL between and prior.
- Advantages over AE: probabilistic, calibrated uncertainty, generative — can synthesize plausible normals.
- Weakness: still needs threshold tuning; failure mode: 'likelihood > uniform' issue (in-distribution mismatch on OOD data — Nalisnick et al. 2019).
Check yourself — multiple choice
- Same as AE
- VAE: reconstruction error or negative ELBO or as anomaly score; probabilistic + calibrated; watch Nalisnick likelihood pitfall on OOD
- Random
- Not real
VAE anomaly: ELBO / recon error / KL; watch Nalisnick OOD pitfall.
#anomaly-detection#deep-learning
Practise Unsupervised Learning
214 interview questions in this topic.