1: \begin{abstract}

2: Stochastic Gradient Langevin Dynamics (SGLD) is a poweful algorithm for optimizing a non-convex objective, where 

3: a controlled and properly scaled Gaussian noise is added to the stochastic gradients to steer the iterates towards a global minimum. SGLD is based on the overdamped Langevin diffusion which is reversible in time. 

4: By adding an anti-symmetric matrix to the drift term of the overdamped Langevin diffusion, one gets a non-reversible diffusion

5: that converges to the same stationary distribution

6: with a faster convergence rate. 

7: In this paper, we study the Mon-reversible Stochastic Gradient Langevin Dynamics (NSGLD) which is based on discretization of the non-reversible Langevin diffusion.

8: We provide finite-time performance bounds for the global convergence of NSGLD for solving stochastic non-convex optimization problems. Our results lead to non-asymptotic guarantees for both population and empirical risk minimization problems. Numerical experiments

9: for Bayesian independent component analysis and neural network models

10: show that NSGLD can

11: outperform SGLD

12: with proper choices of the anti-symmetric matrix.

13: \end{abstract}

14: