Search papers, labs, and topics across Lattice.
Department of Computer Science, University of Helsinki, Finland
2
0
2
By mapping probabilities to a Euclidean space, this new Gaussian Process model slashes latent dimensions and achieves calibrated multi-class classification without relying on approximations.
Even with imperfect proxies for latent confounders, robust predictors can still be uniquely identified across domains, provided those domains are sufficiently diverse in their latent structure.
