2 problems

For Bayesian models such as deep Gaussian processes, the paper distinguishes fractional posteriors with likelihood power $\rho<1$ from the ordinary posterior with $\rho=1$. Ordinar…

Random evolving neural tangent kernel conjecture. In the MoGP case, when the Lévy measure is non-trivial, the infinite-width limit should yield a random kernel, and this kernel sho…