(Master of Information Science) Machine Learning
- Probability Vector: A vector whose elements sum up to one.
- Q. Given two probability vectors r and c (which may not have the same dimensions), how can we manipulate r with minimum cost to transform it into c?
- The minimum cost operation is apparently called the Wasserstein distance between the vectors r and c.
- Sinkhorn algorithm
- “Balancing of matrices”
- Usefulness:
- Allows comparison of images of different sizes (i.e., vectors of different dimensions).
- By using the centroid, we can obtain something like an intermediate image between two images.
- Allows comparison of images of different sizes (i.e., vectors of different dimensions).