1: \begin{abstract}                          % Abstract of not more than 200 words.

2: Privacy protection  and nonconvexity are two challenging problems in

3: decentralized optimization and learning involving sensitive data.

4: Despite some recent advances  addressing each of the two problems

5: separately, no results have been reported that  have theoretical

6: guarantees on both privacy protection and saddle/maximum avoidance

7: in decentralized nonconvex optimization. We propose a new algorithm

8: for decentralized nonconvex optimization that can enable both

9: rigorous differential privacy and   saddle/maximum avoiding

10: performance. The new algorithm allows the incorporation of

11: persistent additive noise to enable rigorous differential privacy

12: for  data samples, gradients, and intermediate optimization

13: variables without losing provable convergence, and thus

14: circumventing the dilemma of trading accuracy for privacy in

15: differential privacy design. More interestingly, the

16:  algorithm is  theoretically proven to be able to  efficiently { guarantee accuracy by avoiding}  convergence to local

17: maxima and saddle points, which has not been reported before in the

18: literature on decentralized nonconvex optimization. The algorithm is

19: efficient in both communication (it only shares one variable in each

20: iteration) and computation (it is encryption-free), and hence is

21: promising for large-scale nonconvex optimization and learning

22:  involving high-dimensional optimization parameters. Numerical experiments

23:  for

24: both a decentralized estimation problem and an Independent Component

25: Analysis (ICA) problem

26:  confirm the effectiveness of the proposed approach.

27: \end{abstract}