This paper introduces and analyzes a search and retrieval model that adopts key semantic communication principles from retrieval-augmented generation. We specifically present an information-theoretic analysis of a remote document retrieval system operating over a symbol erasure channel. The proposed model encodes the feature vector of a query, derived from term-frequency weights of a language corpus by using a repetition code with an adaptive rate dependent on the contextual importance of the terms. At the decoder, we select between two documents based on the contextual closeness of the recovered query. By leveraging a jointly Gaussian approximation for both the true and reconstructed similarity scores, we derive an explicit expression for the retrieval error probability, i.e., the probability under which the less similar document is selected. Numerical simulations on synthetic and real-world data (Google NQ) confirm the validity of the analysis. They further demonstrate that assigning greater redundancy to critical features effectively reduces the error rate, highlighting the effectiveness of semantic-aware feature encoding in error-prone communication settings.

Distributed computations, such as distributed matrix multiplication, can be vulnerable to significant security issues, notably Byzantine attacks. These attacks may target either worker nodes or servers, potentially leading to faulty results that can significantly degrade the overall performance. Therefore, detecting Byzantine attackers and mitigating their effects are crucial in such systems. Motivated by the goal of establishing a secure decentralized matrix-multiplication system, we first introduce a verification method named Common Tag, inspired by the well-known Freivalds’ algorithm, able to verify the multiplication results independent of their associated input matrices. Then, we propose two schemes for sparse matrix multiplication where a group of nodes collaboratively performs a computation task over a logical ring. We consider a subset of Byzantine nodes in the system that may arbitrarily corrupt either their result or any other result passing through them. In Scheme I considering the highly sparse nature of input matrices, we assume that each node has sufficient capacity to store the entire input matrices, and the nodes forward the read-only versions of their computed blocks so that other nodes cannot corrupt them. In Scheme II, we relax the above assumptions, firstly, by considering a limited storage capacity for each node. Secondly, we introduce more powerful adversaries capable of corrupting other nodes’ results by relaxing the read-only assumption. The results demonstrate the feasibility of both schemes and show a significant improvement in terms of distortion over the case where no detection happens. The results also provide a trade-off between the computational complexity required at each node and the reconstruction distortion in both schemes.

We introduce the paradigm of validated decentralized learning for undirected networks with heterogeneous data and possible adversarial infiltration. We require ($a$) convergence to a global empirical loss minimizer when adversaries are absent, and$(\boldsymbol{b})$either detection of adversarial presence or convergence to an admissible consensus model in their presence. This contrasts sharply with the traditional byzantine-robustness requirement of convergence to an admissible consensus irrespective of the adversarial configuration. To this end, we propose the Valid protocol which, to the best of our knowledge, is the first to achieve a validated learning guarantee. Moreover, Valid offers an$O(1/T)$convergence rate (under pertinent regularity assumptions), and computational and communication complexities comparable to non-adversarial distributed stochastic gradient descent. Remarkably, Valid retains optimal performance metrics in adversary-free environments, sidestepping the robustness penalties observed in prior byzantine-robust methods. A distinctive aspect of our study is a heterogeneity metric based on the norms of individual agents' gradients computed at the global empirical loss minimizer. This not only provides a natural statistic for detecting significant byzantine disruptions but also allows us to prove the optimality of Valid in wide generality. Lastly, our numerical results reveal that, in the absence of adversaries, Validcon-verges faster than state-of-the-art byzantine robust algorithms, while when adversaries are present, Valid terminates with each honest agent either converging to an admissible consensus or declaring adversarial presence in the network.

In this paper, we propose a sparse matrix multiplication in a decentralized setting, where a set of worker nodes wishes to compute a task collaboratively over a logical ring. We consider a subset of Byzantine nodes in the system who want to maliciously corrupt the result by corrupting their own computed blocks. In particular, the main focus of this paper is to compute the result with the least possible distortion by identifying the Byzantine nodes and re-assigning their tasks to the benign nodes. Our results demonstrate the feasibility of our proposed decentralized scheme and provide a trade-off between the computational complexity required at each worker node and the reconstruction distortion.

In this paper, we investigate simultaneous lightwave information and power transfer (SLIPT) for underwater visible light communication systems. We consider three SLIPT methods namely time switching (TS), power splitting (PS) and time switching-power splitting (TS-PS) where the splitting/switching factors are defined as optimization parameters. For each of these methods, we derive closed-form expressions for the average harvested energy, bit error rate and spectral efficiency in the presence of underwater turbulence modeled by lognormal statistics. Using these expressions, we determine the optimal splitting factors to maximize the harvested energy while satisfying a given bit error rate value and a given threshold spectral efficiency value. Our results reveal that, if not optimized, SLIPT methods under consideration are outperformed by the simple AC-DC separation (ADS) method which provides the largest harvested energy versus spectral efficiency (HE-SE) region. Optimization of splitting/switching factors extends the HE-SE regions; hence, optimized versions of TS, PS and TS-PS methods are able to significantly outperform ADS for most cases. We further investigate the effect of various channel and system parameters such as water type, turbulence level, beam divergence, receiver aperture size on the harvested energy and quantify the improvements in battery lifetime through the use of SLIPT methods.

Visible light communication (VLC) has emerged as a high-capacity connectivity solution for underwater sensor networks. Since water is relatively transparent to blue or green light, visible light lasers or LEDs can be used as transmitters for underwater wireless connectivity with data rates up to hundreds of Mbps. In underwater networks, a critical system design issue is the network lifetime which highly depends on the battery capacity. Since recharging in underwater scenarios is typically very costly and impractical, energy harvesting can be considered as a promising alternative. In this paper, we explore simultaneous lightwave information and power transfer (SLIPT) for VLC-based USNs. We adopt time splitting method where the receiver switches in time between the modes of energy harvesting (EH) and information decoding (ID). We derive a closed-form expression for the average harvested energy over log-normal model underwater turbulence channel. Using this expression, we determine the splitting factor between EH and ID operation modes to maximize the harvested energy while satisfying a given bit error rate value.