1: \begin{abstract}

2: Stochastic gradient Langevin dynamics (SGLD) is a computationally efficient sampler for Bayesian posterior inference given a large scale dataset.

3: Although SGLD is designed for unbounded random variables, many practical models incorporate variables with boundaries such as non-negative ones or those in a finite interval.

4: To bridge this gap, we consider mapping unbounded samples into the target interval.

5: This paper reveals that several mapping approaches commonly used in the literature produces erroneous samples from theoretical and empirical perspectives. 

6: We show that the change of random variable using an invertible Lipschitz mapping function overcomes the pitfall as well as attains the weak convergence.

7: Experiments demonstrate its efficacy for widely-used models with bounded latent variables including Bayesian non-negative matrix factorization and binary neural networks.

8: \end{abstract}

9: