Top-K Ranking Deep Contextual Bandits for Information Selection Systems

Optimal Transport for Super Resolution Applied to Astronomy Imaging

“…This is relevant for correcting distortions and noise in imaging which we showed for star cluster detection. Another benefit of a predictive model for star clusters is that it can produce a policy that informs where future surveys should look for star clusters [7], [18].…”

“…Problem ( 7) and ( 8) are essentially dual, and for generic data ν there is a mapping λ → λ ′ (λ) such that µ is a solution to (7) if and only if µ is a solution to (8). We will approach (8) from a theoretical perspective in Section III but use (7) in our application since it fits well into a gradient descent method. At least one minimizer exists by compactness of the finite dimensional probability simplex.…”

“…At least one minimizer exists by compactness of the finite dimensional probability simplex. The entropy term in (7) favors sparse solutions. Naturally, there is a trade-off between sparsity of the solution and proximity to the measurement.…”