1: \begin{abstract}

2: Large scale Bayesian nonparametrics (BNP) learner such as stochastic

3: variational inference (SVI) can handle datasets with large class number

4: and large training size at fractional cost. Like its predecessor,

5: SVI rely on the assumption of conjugate variational posterior to approximate

6: the true posterior. A more challenging problem is to consider large

7: scale learning on non-conjugate posterior. Recent works in this direction

8: are mostly associated with using Monte Carlo methods for approximating

9: the learner. However, these works are usually demonstrated on non-BNP

10: related task and less complex models such as logistic regression,

11: due to higher computational complexity. In order to overcome the issue

12: faced by SVI, we develop a novel approach based on the recently proposed

13: variational maximization-maximization (VMM) learner to allow large

14: scale learning on non-conjugate posterior. Unlike SVI, our VMM learner

15: does not require closed-form expression for the variational posterior

16: expectatations. Our only requirement is that the variational posterior

17: is differentiable. In order to ensure convergence in stochastic settings,

18: SVI rely on decaying step-sizes to slow its learning. Inspired by

19: SVI and Adam, we propose the novel use of decaying step-sizes on both

20: gradient and ascent direction in our VMM to significantly improve

21: its learning. We show that our proposed methods is compatible with

22: ResNet features when applied to large class number datasets such as

23: MIT67 and SUN397. Finally, we compare our proposed learner with several

24: recent works such as deep clustering algorithms and showed we were

25: able to produce on par or outperform the state-of-the-art methods

26: in terms of clustering measures.

27: 

28: %\footnote{K. Lim and X. Jiang are with the Rapid-Rich Object Search lab, School

29: %of Electrical and Electronic Engineering, Nanyang Technological University,

30: %Singapore 639798 (email: lkartl@yahoo.com.sg, exdjiang@ntu.edu.sg)}

31: \end{abstract}