Implementation of various generative neural network models for anomaly detection in Julia, using the Flux framework. Serves as a codebase for the comparative study presented in the paper
Škvára, Vít, Tomáš Pevný, and Václav Šmídl. Are generative deep models for novelty detection truly better? arXiv preprint arXiv:1807.05027 (2018).
| acronym | name | paper |
|---|---|---|
| AE | Autoencoder | Vincent, Pascal, et al. "Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion." Journal of Machine Learning Research 11.Dec (2010): 3371-3408. link |
| VAE | Variational Autoencoder | Kingma, Diederik P., and Max Welling. "Auto-encoding variational bayes." arXiv preprint arXiv:1312.6114 (2013). link |
| sVAE | symetric Variational Autoencoder | Pu, Yunchen, et al. "Symmetric variational autoencoder and connections to adversarial learning." arXiv preprint arXiv:1709.01846 (2017). link |
| GAN | Generative Adversarial Network | Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014. link |
| fmGAN | GAN with feature-matching loss | Salimans, Tim, et al. "Improved techniques for training gans." Advances in Neural Information Processing Systems. 2016. link |
Experiments are executed on the Loda (Lightweight on-line detector of anomalies) datasets that are in the experiments directory. The sampling method is based on this paper. After downloading the datasets, you can create your own using the experiments/prepare_data.jl function. For experimental evaluation, you need the EvalCurves package:
>> Pkg.clone("https://github.com/vitskvara/EvalCurves.jl.git")