Statistical Translation, Heat Kernels and Expected Distances
This work addresses the challenge of poor representation in statistical modeling for text and image data, offering a novel approach to metric learning.
The authors tackled the problem of representing high-dimensional structured data like text and images, which is often misrepresented in statistical modeling, by developing a new framework for unsupervised metric learning based on connections between statistical translation, heat kernels, and expected distances, resulting in distances that are generally superior to standard counterparts.
High dimensional structured data such as text and images is often poorly understood and misrepresented in statistical modeling. The standard histogram representation suffers from high variance and performs poorly in general. We explore novel connections between statistical translation, heat kernels on manifolds and graphs, and expected distances. These connections provide a new framework for unsupervised metric learning for text documents. Experiments indicate that the resulting distances are generally superior to their more standard counterparts.