Dear Mahesh,

Would you please give us more details in what you are trying to achieve? Learning in Unsupervised suggest an already classified dataset.

Yohann

Hi,

Depends on your data, you can find what kind of clustering algorithm is proper to solve your problem. However, you can search for Deep Embedded Clustering (DEC) which is one of the most promising approaches in this matter.

Good Luck.

Saeed

Dear Mahesh,

In addition to Saeed's post, Deep Embedding may be declined in 2 versions:

• an unsupervised one (DEC): Xie et al., " Unsupervised Deep Embedding for Clustering Analysis ", 2016 - https://arxiv.org/pdf/1511.06335.pdf
• a supervised one (DEK): Le et al., " Deep Embedding Kernel ", 2018 - https://arxiv.org/pdf/1804.05806.pdf

More generally, Deep Kernel Processes match your needs. I suggest that you follow:

• Lu et al., " How to Scale Up Kernel Methods to Be As Good As Deep Neural Nets ", 2015 - https://arxiv.org/pdf/1411.4000.pdf
• Anselmi et al., " Deep Convolutional Networks are Hierarchical Kernel Machines ", 2015 - https://pdfs.semanticscholar.org/7227/d4e427c89202e4368922bb7f0304ee82a582.pdf
• Wilson et al., " Gaussian Process Kernels for Pattern Discovery and Extrapolation ", 2013 - https://arxiv.org/pdf/1302.4245.pdf
• Wilson et al., " Thoughts on Massively Scalable Gaussian Processes ", 2015 - https://arxiv.org/pdf/1511.01870.pdf
• Montavon et al., " Kernel Analysis of Deep Networks ", 2011 - http://jmlr.csail.mit.edu/papers/volume12/montavon11a/montavon11a.pdf
• Wilson et al., " Deep Kernel Learning ", 2016 - http://proceedings.mlr.press/v51/wilson16.pdf
• Song et al., " Optimizing Kernel Machines using Deep Learning ", 2017 - https://arxiv.org/pdf/1711.05374.pdf
• Bengio et al., " Representation Learning: A Review and New Perspectives ", 2014 - https://arxiv.org/pdf/1206.5538.pdf

Hoping it will be helpful to you,

Regards

Respected Yohann Missiak and Saeed Khazaee,

Thanks for your valuable reply...

I have an image data,

from which I want to identify the anomalies in an unsupervised passion using deep learning model.

In this regard,

I need the unsupervised deep learning model..

Anomaly detection may be addressed by means of Pattern Discovery-derived solutions (refer to my prior post)...

Respected Stéphane Breton ,

Thanks for your guidance and detailed references..

I will go through these links...

Stéphane Breton Sir,

Thank you for your reply...

I want to identify the specific image pixels using unsupervised deep learning model.

For this sort of unsupervised work, which unsupervised deep learning model will be effective?

Dear Mahesh,

Your specificity criterion largely depends on your application context. Nevertheless, you may check out:

• Kiran et al., " An overview of deep learning based methods for unsupervised and semi-supervised anomaly detection in videos ", 2018 - https://arxiv.org/pdf/1801.03149.pdf
• Verdoja et al., " Graph Laplacian for Image Anomaly Detection ", 2018 - https://arxiv.org/pdf/1802.09843.pdf
• Kwon et al., " Kernel RX-Algorithm: A Nonlinear Anomaly Detector for Hyperspectral Imagery ", 2005 -

  Article Kernel RX-Algorithm: A Nonlinear Anomaly Detector for Hypers...

• Reece et al., " Anomaly Detection and Removal Using Non-Stationary Gaussian Processes ", 2015 - https://arxiv.org/pdf/1507.00566.pdf
• Herlands et al., " Gaussian Process Subset Scanning for Anomalous Pattern Detection in Non-iid Data ", 2018 - https://arxiv.org/pdf/1804.01466.pdf

Regards

I agree with Stéphane Breton .

Please provide a general explanation of what you are trying to achieve.

For any abnormality detection (considering the picture is not altered), you still need the model to learn what is considered normal or not.

Unsupervised model would only classify images into specific group the model found. It does not mean it is abnormal, it just mean there are trends in the image that made them similar with each other.

If you want to find altered pictures, then you need different other tools.