We study differentially private stochastic optimization in convex and non-convex settings. For the convex case, we focus on the family of non-smooth generalized linear losses (GLLs). Our algorithm for the $\ell_2$ setting achieves optimal excess population risk in near-linear time, while the best known differentially private algorithms for general convex losses run in super-linear time.

Its structure and organization could be substantially improved. The notation is unclear, and the terminology is not defined. For example, see lines 45-48. The formal problem statement (section 2.2) is vague, as well. Many technical terms are used without any context; they are not explained and, further, it is not clear how those concepts help the claims made in the paper.

In this paper, the authors extend the "resource allocation with semi-bandit feedback", proposed by Lattimore et al. [2014], to the multi-resource case. The paper has provided two regret bounds, one for the worst case (Theorem 2) and the other for the "resource-laden" case (Theorem 7). The authors also provide a new result on the "weighted least squares estimation", which is independently interesting. The paper is well-written and very interesting, the analysis in this paper is also rigorous. The extension to the multi-resource case is non-trivial, and the new result on the "weighted least squares estimation" is very interesting and might be reused by researchers in the field of bandit/RL in the future. Thus, I think this paper meets the acceptance threshold.