Coresets: 8 Future Directions
Several future ideas to work upon are the following,
Can we decompose a kernel \(K\) into feature matrix \(X\) to leverage it into linear regression coreset algorithm? This way we would be creating a coreset for any Gaussian process models in general.
Practical comparison (time-complexity, performance, number of coreset points) of coresets created by classical algorithms and active learning strategies.