We study differential privacy in a distributed setting where two parties would like to perform analysis of their joint data while preserving privacy for both datasets. Our results imply almost tight lower bounds on the accuracy of such data analyses, both for specific natural functions (such as Hamming distance) and in general. Our bounds expose a sharp contrast between the two-party setting and the simpler client-server setting (where privacy guarantees are one-sided). In addition, those bounds demonstrate a dramatic gap between the accuracy that can be obtained by differentially private data analysis versus the accuracy obtainable when privacy is relaxed to a computational variant of differential privacy. The first proof technique we develop demonstrates a connection between differential privacy and deterministic extraction from Santha-Vazirani sources. A second connection we expose indicates that the ability to approximate a function by a low-error differentially private protocol is strongly related to the ability to approximate it by a low communication protocol. (The connection goes in both directions.)

The Limits of Two-Party Differential Privacy.

It does not appear to be widely known that the moment-generating function Mx (t) contains information about negative as well as positive integer moments. Proofs of this are provided, extensions are indicated, and earlier literature deriving and using this type of result is briefly noted. Some examples are given.

https://www.jstor.org/stable/pdfplus/2683982.pdf

The Moment-Generating Function and Negative Integer Moments

/pdf/the-limits-of-two-party-differential-privacy-45917rbiyt.pdf

The Limits of Two-Party Differential Privacy

In networks with large latency, feedback about received packets may lag considerably the transmission of the original packets, limiting the feedback's usefulness. Moreover, time duplex constraints may entail that receiving feedback may be costly. In this work, we consider tailoring feedback and coding jointly in such settings to reduce the expected delay for successful in order reception of packets. We find that, in certain applications, judicious choices provide results that are close to those that would be obtained with a full-duplex system. We study two cases of data transmission: one-to-all broadcast and all-to-all broadcast. We also analyze important practical considerations weighing the trade off between performance and complexity in applications that rely on random linear network coding. Finally, we study the problem of transmission of information under the large latency and time duplexing constraints in the presence of random packet arrivals. In particular, we analyze the problem of using a batch by batch approach and an online network coding approach with Poisson arrivals. We present numerical results to illustrate the performance under a variety of scenarios and show the benefits of the proposed schemes as compared to typical ARQ and scheduling schemes.

/pdf/on-coding-for-delay-network-coding-for-time-division-kjyk3c8xk8.pdf

On Coding for Delay—Network Coding for Time-Division Duplexing

Motivated by the many real-world applications of reinforcement learning (RL) that require safe-policy iterations, we consider the problem of off-policy evaluation (OPE) --- the problem of evaluating a new policy using the historical data obtained by different behavior policies --- under the model of nonstationary episodic Markov Decision Processes (MDP) with a long horizon and a large action space. Existing importance sampling (IS) methods often suffer from large variance that depends exponentially on the RL horizon $H$. To solve this problem, we consider a marginalized importance sampling (MIS) estimator that recursively estimates the state marginal distribution for the target policy at every step. MIS achieves a mean-squared error of $$ \frac{1}{n} \sum_{t=1}^H\mathbb{E}_{\mu}\left[\frac{d_t^\pi(s_t)^2}{d_t^\mu(s_t)^2} \Var_{\mu}\left[\frac{\pi_t(a_t|s_t)}{\mu_t(a_t|s_t)}\big( V_{t+1}^\pi(s_{t+1}) + r_t\big) \middle| s_t\right]\right] + \tilde{O}(n^{-1.5}) $$ where $\mu$ and $\pi$ are the logging and target policies, $d_t^{\mu}(s_t)$ and $d_t^{\pi}(s_t)$ are the marginal distribution of the state at $t$th step, $H$ is the horizon, $n$ is the sample size and $V_{t+1}^\pi$ is the value function of the MDP under $\pi$. The result matches the Cramer-Rao lower bound in [Jiang and Li, 2016] up to a multiplicative factor of $H$. To the best of our knowledge, this is the first OPE estimation error bound with a polynomial dependence on $H$. Besides theory, we show empirical superiority of our method in time-varying, partially observable, and long-horizon RL environments.

/pdf/towards-optimal-off-policy-evaluation-for-reinforcement-nup2itu1lf.pdf

Towards Optimal Off-Policy Evaluation for Reinforcement Learning with Marginalized Importance Sampling

We investigate the problem of finding the expected value of functions of a random variable X of the form f(X) = (X+A)−n where X+A>0 a.s. and n is a non-negative integer. The technique is to successively integrate the probability generating function and is suggested by the well-known result that successive differentiation leads to the positive moments. The technique is applied to the problem of finding E[1/(X+A)] for the binomial and Poisson distributions.

https://www.jstor.org/stable/pdfplus/2284399.pdf

M. T. Chao

Papers

Negative Moments of Positive Random Variables