unsupervised_anomaly_detection

Existing Approaches

We can classify the existing approaches for "unsupervised" anomaly/outlier detection based on (at least) two criterias:

• The "contamination" of the training dataset

  Based on this criteria, each approach will assume one of the following:

  • Contaminated dataset: the training dataset contains both normal and anamolous elements.
  • Uncontaminated dataset: The training dataset contains only normal samples. Such approach is also refered asweakly supervised.

• How far is the distribution of the anomalous samples to the normal samples

  • If the anamolous samples lie far away from the normal distribution (as shown in (b)), the problem is known is as out of distribution detection (OoDD) . Note that OoDD is sometimes refered as a seperate problem and not included in the family of anomaly deteion, but here I will assume that the OoDD detection problem lies under the anomaly detection umberlla.

  • On the other hand, if the anamolous samples lie at the boundaries of normal distribution (as shown in (b)), I will refer to the problem as fault detection (FD).

Note that to the best of my knowledge, there is no existing approach that tackles the problem of fault detection(FD)assuming a contaminated dataset.

UFDGAN

Assumptions:

• Contaminated training set
• Fault detection (not OoDD)
• The fault percentage is known

Why the problem is quasi unsolvobale even for a human?

let's take the MTSD dataset as an example. As shown in the bellow figure, (a) is a sample of the majority normal samples, (b) is a sample of a minority normal data and (c) is a sample of a the faulty samples (also a minority). So, the dataset is composed of multi modal data, but only one mode is labaled as anomaly.

If the dataset is handed to a human, he most likely will not be able to correctly detect the anomalous samples.

So, I argue that the problem of unsupervised fault detection assuming a contaminated training dataset is challenging even for a human being. And in order to be solvoble, the data distribution should be exactly a mixture of ONLY two salient modes, which may rarely be the case in the real world.

Out Of Distribution Detection through GAN's Mode Collapse

We adopt the same assumptions as the UFDGAN paper, but we tackle the problem of Out OF Distribution detection.

Recent Related Work

SSD: A unified framework for selfsupervised outlier detection..

CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances.

Approach

The model is composed of a GAN (generator and a discriminator), along with an encoder. and the encoder tries to learn the reverse mapping from the image space to the gan's latent space. After each iteration, based on the reconstruction error, the is filtered out.

Dataset

The MNIST dataset is used. The 0 is used as the normal and sampled from the different orher classes. the anomaly percentage is 5%.