Because the exchange of information among all the users in a large network can take a long time, a successive omniscience protocol is proposed. Namely, subgroups of users first recover the information of other users in the same subgroup at an earlier stage called local omniscience. Then, the users recover the information of all other users at a later stage called global omniscience. To facilitate the information exchange, a distributed storage system is used, so that users can conveniently upload and download messages through some reliable central servers. The minimum upload bandwidth is characterized and a bandwidth-storage trade-off is discovered. The results reveal the new connections to the problem of secret key agreement and, consequently, provide meaningful interpretations of a recently proposed multivariate mutual information measure that was inspired by the secret key agreement problem.
This paper investigates lift, the likelihood ratio between the posterior and prior belief about sensitive features in a dataset. Maximum and minimum lifts over sensitive features quantify the adversary’s knowledge gain and should be bounded to protect privacy. We demonstrate that max- and min-lifts have a distinct range of values and probability of appearance in the dataset, referred to as lift asymmetry. We propose asymmetric local information privacy (ALIP) as a compatible privacy notion with lift asymmetry, where different bounds can be applied to min- and max-lifts. We use ALIP in the watchdog and optimal random response (ORR) mechanisms, the main methods to achieve lift-based privacy. It is shown that ALIP enhances utility in these methods compared to existing local information privacy, which ensures the same (symmetric) bounds on both max- and min-lifts. We propose subset merging for the watchdog mechanism to improve data utility and subset random response for the ORR to reduce complexity. We then investigate the related lift-based measures, including -norm, -privacy criterion, and -lift. We reveal that they can only restrict max-lift, resulting in significant min-lift leakage. To overcome this problem, we propose corresponding lift-inverse measures to restrict the min-lift. We apply these lift-based and lift-inverse measures in the watchdog mechanism. We show that they can be considered as relaxations of ALIP, where a higher utility can be achieved by bounding only average max- and min-lifts.
9 October 2017 in IEEE Transactions on Information Theory
This paper considers the communication for omniscience problem: a set of users observe a discrete memoryless multiple source and want to recover the entire multiple source via noise-free broadcast communications. We study the problem of how to determine an optimal rate vector that attains omniscience with the minimum sum rate, the total number of communications. The results cover both asymptotic and non-asymptotic models where the transmission rates are real and integral, respectively. We propose a modified decomposition algorithm (MDA) and a sum-rate increment algorithm (SIA) for the asymptotic and non-asymptotic models, respectively, both of which determine the value of the minimum sum rate and a corresponding optimal rate vector in polynomial time. For the coordinate saturation capacity algorithm, a nesting algorithm in MDA and SIA, we propose to implement it by a fusion method and show by experimental results that this fusion method contributes to a reduction in computation complexity. Finally, we show that the separable convex minimization problem over the optimal rate vector set in the asymptotic model can be decomposed by the fundamental partition, the optimal partition of the user set that determines the minimum sum rate, so that the problem can be solved more efficiently.
None of the features commonly utilised in automatic emotion classification systems completely disassociate emotion-specific information from speaker-specific information. Consequently, this speaker-specific variability adversely affects the performance of the emotion classification system and in existing systems is frequently mitigated by some form of speaker normalisation. Speaker adaptation offers an alternative to normalisation and this paper proposes a novel bootstrapping technique which involves selecting appropriate initial models from a large training pool, prior to speaker adaptation of emotion models in the context of GMM based emotion classification as an alternative to speaker normalisation. Evaluations on the LDC Emotional Prosody and the FAU Aibo corpora reveal that an emotion classification system based on the proposed bootstrapping method outperforms systems based on speaker normalisation as long as a small amount of labelled adaptation data is available. It also outperforms speaker adaption from common initial models estimated from all training speakers.
We propose a coalition game model for the problem of communication for omniscience (CO). In this game model, the core contains all achievable rate vectors for CO with sum-rate being equal to a given value. Any rate vector in the core distributes the sum-rate among users in a way that makes all users willing to cooperate in CO. We give the necessary and sufficient condition for the core to be nonempty. Based on this condition, we derive the expression of the minimum sum-rate for CO and show that this expression is consistent with the results in multivariate mutual information (MMI) and coded cooperative data exchange (CCDE). We prove that the coalition game model is convex if the sum-rate is no less than the minimal value. In this case, the core is non-empty and a rate vector in the core that allocates the sum-rate among the users in a fair manner can be found by calculating the Shapley value.
We consider the problem of finding the minimum sum-rate strategy in cooperative data exchange systems that do not allow packet-splitting (NPS-CDE). In an NPS-CDE system, there are a number of geographically close cooperative clients who send packets to help the others recover a packet set. A minimum sum-rate strategy is the strategy that achieves universal recovery (the situation when all the clients recover the whole packet set) with the the minimal sum-rate (the total number of transmissions). We propose an iterative merging (IM) algorithm that recursively merges client sets based on a lower estimate of the minimum sum-rate and updates to the value of the minimum sum-rate. We also show that a minimum sum-rate strategy can be learned by allocating rates for the local recovery in each merged client set in the IM algorithm. We run an experiment to show that the complexity of the IM algorithm is lower than that of the existing deterministic algorithm when the number of clients is lower than 94.