Timezone: »

 
Poster
Time-independent Generalization Bounds for SGLD in Non-convex Settings
Tyler Farghly · Patrick Rebeschini

Thu Dec 09 12:30 AM -- 02:00 AM (PST) @
in Poster Session 5 »

We establish generalization error bounds for stochastic gradient Langevin dynamics (SGLD) with constant learning rate under the assumptions of dissipativity and smoothness, a setting that has received increased attention in the sampling/optimization literature. Unlike existing bounds for SGLD in non-convex settings, ours are time-independent and decay to zero as the sample size increases. Using the framework of uniform stability, we establish time-independent bounds by exploiting the Wasserstein contraction property of the Langevin diffusion, which also allows us to circumvent the need to bound gradients using Lipschitz-like assumptions. Our analysis also supports variants of SGLD that use different discretization methods, incorporate Euclidean projections, or use non-isotropic noise.

Keywords: Optimization

Author Information

Tyler Farghly (University of Oxford)
Patrick Rebeschini (University of Oxford)

More from the Same Authors