This paper studies federated learning (FL) with non-identical data generating distributions and under the notion of proportional fairness. Users are partitioned into disjoint clusters based on their data distributions, and a distinct model is learnt for each cluster. Different than previous work that considered clustering to optimize personalized learning performance, here clustering is inspected under the disparate impact doctrine which requires that protected classes (e.g., those of certain race or gender) have representations in every cluster that are approximately equal to their representation in the overall set of users. A family of iterative algorithms that balance the learning performance and proportional fairness through cluster assignments as randomized functions of the learning losses is proposed. The trade-off induced by our algorithms between accuracy of cluster estimation and the introduced randomization level is characterized. The proposed algorithm is examined on a real dataset to evaluate its performance.

We focus on causal discovery in the presence of measurement error in linear systems where the mixing matrix, i.e., the matrix indicating the independent exogenous noise terms pertaining to the observed variables, is identified up to permutation and scaling of the columns. We demonstrate a somewhat surprising connection between this problem and causal discovery in the presence of unobserved parentless causes, in the sense that there is a mapping, given by the mixing matrix, between the underlying models to be inferred in these problems. Consequently, any identifiability result based on the mixing matrix for one model translates to an identifiability result for the other model. We characterize to what extent the causal models can be identified under a two-part faithfulness assumption. Under only the first part of the assumption (corresponding to the conventional definition of faithfulness), the structure can be learned up to the causal ordering among an ordered grouping of the variables but not all the edges across the groups can be identified. We further show that if both parts of the faithfulness assumption are imposed, the structure can be learned up to a more refined ordered grouping. As a result of this refinement, for the latent variable model with unobserved parentless causes, the structure can be identified. Based on our theoretical results, we propose causal structure learning methods for both models, and evaluate their performance on synthetic data.

In this paper, new two-user multiple access wiretap channel models are studied. First, the multiple access wiretap channel II with a discrete memoryless main channel under different wiretapping scenarios is introduced. The wiretapper, as in the classical wiretap channel II model, chooses a fixed-size subset of the channel uses, in which it obtains noise-free observations of one of the codewords: a deterministic function, e.g., superposition, of the two codewords or each of the two codewords. A fourth wiretapping scenario is considered, in which the wiretapper, in each position it chooses, decides to observe either one of the codewords or both codewords, with an overall budget on the number of its noiselessly observed symbols. These, thus, extend the recently examined wiretap channel II with a noisy main channel to a multiple access setting with a variety of attack models for the wiretapper. Next, the proposed multiple access wiretap channel II models are further generalized to the case when the wiretapper observes the outputs of a discrete memoryless channel, instead of erasures, outside the subset of noiseless observations. Achievable strong secrecy rate regions for all the proposed models are derived. Achievability is established by solving dual multi-terminal secret key agreement problems in the source model and converting the solution to the original channel models using probability distribution approximation arguments. The derived achievable rate regions quantify the secrecy cost due to the additional capabilities of the wiretapper with respect to the previous multiple access wiretap models.

This paper introduces the notion of cache-tapping into information theoretic models of coded caching. In particular, the wiretap II model with two receivers equipped with fixed-size cache memories is considered. The adversary chooses a set of symbols from cache placement, delivery, or both to tap into. The legitimate parties know neither whether cache placement, delivery, or both transmissions are tapped, nor the positions of tapped symbols. Only the size of overall tapped set is known. The strong secrecy capacity, i.e., the maximum achievable file rate while keeping the overall library strongly secure, is identified for the instance of two library files. Achievability is established using a code design which combines wiretap coding, security embedding codes, one-time pad keys, and coded caching. The study overall demonstrates that information theoretic security guarantees are possible against a powerful adversary which optimizes its attack over both phases of a cache-aided communication system.

In this paper, a new wiretap channel model is proposed, where the legitimate transmitter and receiver communicate over a discrete memoryless channel. The wiretapper has perfect access to a fixed-length subset of the transmitted code word symbols of her choosing. Additionally, she observes the remainder of the transmitted symbols through a discrete memoryless channel. This new model subsumes the classical wiretap channel and wiretap channel II with noisy main channel as its special cases, and is termed as the generalized wiretap channel for that reason. The strong secrecy capacity of the proposed channel model is identified. Achievability is established by solving a dual secret key agreement problem in the source model, and converting the solution to the original channel model using probability distribution approximation arguments. In the dual problem, a source encoder and decoder, who observe random sequences independent and identically distributed according to the input and output distributions of the legitimate channel in the original problem, communicate a confidential key over a public error-free channel using a single forward transmission, in the presence of a compound wiretapping source who has perfect access to the public discussion. The security of the key is guaranteed for the exponentially many possibilities of the subset chosen at the wiretapper by deriving a lemma which provides a doubly-exponential convergence rate for the probability that, for a fixed choice of the subset, the key is uniform and independent from the public discussion and the wiretapping source's observation. The converse is derived by using Sanov's theorem to upper bound the secrecy capacity of the generalized wiretap channel by the secrecy capacity when the tapped subset is randomly chosen by nature.

A new model for the interference channel with confidential messages (IC-CM) is introduced, where each receiver, besides his noisy observations, is provided with a fixed-length subset, of his choosing, of noiseless observations for the transmitted codewords of both users, making confidential communication more challenging than the previous such model. In addition, in the same spirit, a broadcast channel with confidential messages (BC-CM), where the receivers noiselessly tap into subsets of their choice of the transmitted codeword, is considered. Achievable strong secrecy rate regions for both models are derived. In both models, the size of the subset quantifies a secure rate trade-off between the two receivers. The case of the new BC-CM model with one receiver's noisy observations are degraded with respect to the other receiver, and only the degraded receiver is provided with the subset of noiseless observations, is highlighted. In this case, the receiver with the degraded noisy observations has a positive rate after a certain threshold of his noiseless observations, i.e., with the aid of these symbols.

A new broadcast wiretap channel (B-WTC) with a wiretapper who noiselessly taps into a fixed-length subset of the transmitted symbols of her choice, and observes the remainder through a noisy channel, is studied. An achievable strong secrecy rate region which extends Marton's inner bound to the proposed setting, is derived. Strong secrecy capacity regions for two classes of the new B-WTC, namely the new B-WTC with deterministic receivers, and the new B-WTC with degraded receivers and more noisy wiretapper, are identified. These results extend the recently proposed new wiretap channel model to the broadcast setting.

In this paper, a new wiretap channel (WTC) model with a discrete memoryless (DM) main channel and a wiretapper who noiselessly observes a fixed portion, of her choice, of the transmitted symbols, while observing the remaining transmitted symbols through another DM channel (DMC), is considered. The strong secrecy capacity of the model is identified. The achievability is established using the output statistics of random binning framework which exploits the duality between source and channel coding problems. The converse is derived by upper bounding the secrecy capacity of an equivalent model with the secrecy capacity of a DM-WTC. This result generalizes both the classical DM-WTC and the WTC-II with a DM main channel.

A two transmitter multiple access wiretap channel II (MAC-WT-II) with a discrete memoryless (DM) main channel is investigated. Two models for the wiretapper, who chooses a fixed-length subset of the channel uses and observes erasures outside this subset, are proposed. In the first model, in each position of the subset, the wiretapper noiselessly observes either the first or the second user's symbol, while in the second model, the wiretapper observes a noiseless superposition of the two symbols. Achievable strong secrecy rate regions for the two models are derived. The achievability is established by solving a dual secret key agreement problem in the source model. The secrecy of the keys in the dual source model is established by deriving a lemma which provides a doubly exponential convergence rate for the probability of the keys being uniform and independent from the wiretapper's observation. The results extend the recently examined WTC-II with a DM main channel to a multiple access setting.

A new model for the two-user multiple-access wiretap channel is considered. In this model, the legitimate (main) channel is a discrete memoryless channel (DMC), and the wiretapper chooses a fixed-length subset of the channel uses where she has perfect access to the transmitted symbols of the both users, while observing the remainder of the transmitted codewords through a second DMC. As such, it generalizes the existing multiple-access wiretap channel models and extends the recently proposed new wiretap channel model to a multiple access setting. An achievable strong secrecy rate region for the model is derived. The achievability is established by solving a dual multi-terminal secret key agreement problem in the source model, where two independent sources are communicating confidential keys to a common decoder over a public channel in the presence of a compound wiretapping source. The secrecy of the two keys in the dual source model is established by deriving a lemma which provides a doubly exponential convergence rate for the probability of the keys being uniform and independent from the public discussion and the wiretapping source observation.

The secure degrees of freedom (s.d.o.f.) of a multiantenna Gaussian wiretap channel with N antennas at the transmitter and receiver and an arbitrary number of antennas, M, at the wiretapper is characterized when a multiantenna cooperative jammer (CJ) is available as a helper. This generalizes our previous result that assumed the same number of antennas at the eavesdropper as the legitimate parties. In particular, for arbitrary values of N andM, the s.d.o.f. is derived for all possible values of the number of antenna at the CJ, K. The achievability is based on a variety of signalling, beamforming, and alignment techniques which vary according to the value of K, whether M is larger than, smaller than, or equal to N, and whether the s.d.o.f. is integer valued or not an integer. The converse is based on combining an upper bound for the s.d.o.f. which allows for cooperation between the transmitter and CJ and holds for some values of K, with another upper bound which exploits the secrecy and reliability constraints and holds for other values of K.

In this paper, a wiretap channel where the transmitter and receiver communicate through a discrete memoryless channel, and the eavesdropper (Eve) has perfect access to a fixed fraction of transmitted symbols (of its choosing) is considered. An outer bound for the rate-equivocation region of the channel, for all such fractions, is derived. An achievable scheme, which provides an inner bound for the rate-equivocation region, is proposed. The achievability is established by defining a class of good codebooks for which there exists a good partition that achieves the required level of equivocation no matter what subset of symbols Eve chooses. It is shown that, for a uniform input distribution, the probability of this class of good codes approaches 1 as the block length increases. This generalizes the wiretap II model to one with a noisy main channel.

Online social communities often exhibit complex relationship structures, ranging from close friends to political rivals. As a result, persons are influenced by their friends and foes differently. Network applications can benefit from accompanying these structural differences in propagation schemes. In this paper, we study the optimal influence propagation policies for networks with positive and negative relationship types. We tackle the problem of minimizing the end-to-end propagation cost of influencing a target person in favor of an idea by utilizing the relationship types in the underlying social graph. The propagation cost is incurred by social and physical network dynamics such as frequency of interaction, the strength of friendship and foe ties, propagation delay or the impact factor of the propagating idea. We extend this problem by incorporating the impact of message deterioration and ignorance. We demonstrate our results in both a controlled environment and the Epinions dataset. Our results show that judicious propagation schemes lead to a significant reduction in the average cost and complexity of influence propagation compared to naïve myopic algorithms.

A multiple antenna Gaussian wiretap channel with a multiantenna cooperative jammer (CJ) is considered and the secure degrees of freedom (s.d.o.f.), with N antennas at the sender, receiver, and eavesdropper, is derived for all possible values of the number of antennas at the cooperative jammer, K. In particular, the upper and lower bounds for the s.d.o.f. are provided for different ranges of K and shown to coincide. Gaussian signaling both for transmission and jamming is shown to be sufficient to achieve the s.d.o.f. of the channel, when the s.d.o.f. is integer-valued. By contrast, when the channel has a non-integer s.d.o.f., structured signaling and joint signal space and signal scale alignment are employed to achieve the s.d.o.f.

In this work, we analyze the diversity gain region (DGR) of the single-antenna Rayleigh fading Z-Interference channel (ZIC). More specifically, we characterize the achievable DGR of the fixed-power split Han-Kobayashi (HK) approach under these assumptions. Our characterization comes in a closed form and demonstrates that the HK scheme with only a common message is a singular case, which achieves the best DGR among all HK schemes for certain multiplexing gains. Finally, we show that generalized time sharing, with variable rate and power assignments for the common and private messages, does not improve the achievable DGR.

We characterize the achievable three-dimensional tradeoff between diversity, multiplexing, and delay of the single antenna Automatic Retransmission reQuest (ARQ) Z-interference channel. Non-cooperative and cooperative ARQ protocols are adopted under these assumptions. Considering no cooperation exists, we study the achievable tradeoff of the fixed-power split Han-Kobayashi (HK) approach. Interestingly, we demonstrate that if the second user transmits the common part only of its message in the event of its successful decoding and a decoding failure at the first user, communication is improved over that achieved by keeping or stopping the transmission of both the common and private messages. Under cooperation, two special cases of the HK are considered for static and dynamic decoders. The difference between the two decoders lies in the ability of the latter to dynamically choose which HK special-case decoding to apply. Cooperation is shown to dramatically increase the achievable first user diversity.