Seminars
Home
Centers & Programs
AI and Natural Sciences
Seminars
- FIELD
- AI and Natural Sciences
- DATE
-
Jun 18 (Wed), 2025
- TIME
- 14:00 ~ 16:00
- PLACE
- 7323
- SPEAKER
- Kim, Jakwang
- HOST
- Choi, Jaesung
- INSTITUTE
- University of British Columbia
- TITLE
- Sample amplification by doubly optimal transport
- ABSTRACT
- In this talk, I will give a new method of sample amplification problem. The “sample amplification” problem formalizes the following question: given n i.i.d. coupled samples (X_i, Y_i) drawn by unknown distribution P, and given m i.i.d. marginal samples (X_j) and (Y_j) drawn by marignals P_X and P_Y respectively of P, can we recover coupled samples (X_j, Y_j) from marginal samples to have the same distribution P? A key technique is optimal transport. Setting a proper optimal transport problem, we propose a new method for sample amplification. The duality of the problem is derived, and the qualitative consistency of the proposed estimator is proved by the Gamma-convergence. If time is permitted, remaining technical issues are discussed too. Based on a joint work with Young-Heon Kim(UBC, math) and Chan Park(UIUC, stat).
- FILE
-
