Kernel Information Embeddings

Kernel Information Embeddings (KIE) are probabilistic, non-linear latent variable models.
Shared Kernel Information Embeddings generalize KIE to multiple observation spaces.

Code

This Python module provides an implementation of KIE and shared KIE using GPUs through Tijmen Tieleman's gnumpy package.
It further requires numpy and it can use minimize.py for faster training.

This example-script demonstrates use of the model by instantiating and applying it to this data-set.

References

2006 Memisevic, R.
Kernel Information Embeddings
International Conference on Machine Learning (ICML 2006). [pdf]

2009 Sigal, L., Memisevic, R., Fleet, D.
Shared kernel information embeddings for discriminative inference
Computer Vision and Pattern Recognition (CVPR 2009). [pdf]