Graph Neural Networks (GNNs) have demonstrated remarkable performance in various graph-based machine learning tasks, yet evaluating the importance of neighbors of testing nodes remains largely unexplored due to the challenge of assessing data importance without test labels. To address this gap, we propose Shapley-Guided Utility Learning (SGUL), a novel framework for graph inference data valuation. SGUL innovatively combines transferable data-specific and modelspecific features to approximate test accuracy without relying on ground truth labels. By incorporating Shapley values as a preprocessing step and using feature Shapley values as input, our method enables direct optimization of Shapley value prediction while reducing computational demands. SGUL overcomes key limitations of existing methods, including poor generalization to unseen test-time structures and indirect optimization. Experiments on diverse graph datasets demonstrate that SGUL consistently outperforms existing baselines in both inductive and transductive settings. SGUL offers an effective, efficient, and interpretable approach for quantifying the value of test-time neighbors.

Knowledge Graph Question Answering (KGQA) systems rely on high-quality benchmarks to evaluate complex multi-hop reasoning. However, despite their widespread use, popular datasets such as WebQSP and CWQ suffer from critical quality issues, including inaccurate or incomplete ground-truth annotations, poorly constructed questions that are ambiguous, trivial, or unanswerable, and outdated or inconsistent knowledge. Through a manual audit of 16 popular KGQA datasets—including WebQSP and CWQ—we find that the average factual correctness rate is only 57%. To address these issues, we introduce KGQAGen, an LLM-in-the-loop framework that systematically resolves these pitfalls. KGQAGen combines structured knowledge grounding, LLM-guided generation, and symbolic verification to produce challenging and verifiable QA instances. Using KGQAGen, we construct KGQAGen-10k, a 10K-scale benchmark grounded in Wikidata, and evaluate a diverse set of KG-RAG models. Experimental results demonstrate that even state-of-the-art systems struggle on this benchmark, highlighting its ability to expose limitations of existing models. Our findings advocate for more rigorous benchmark construction and position KGQAGen as a scalable framework for advancing KGQA evaluation.

As an emerging interpretable technique, Generalized Additive Models (GAMs) adopt neural networks to individually learn non-linear functions for each feature, which are then combined through a linear model for final predictions. Although GAMs can explain deep neural networks (DNNs) at the feature level, they require large numbers of model parameters and are prone to overfitting, making them hard to train and scale. Additionally, in real-world datasets with many features, the interpretability of feature-based explanations diminishes for humans. To tackle these issues, recent research has shifted towards concept-based interpretable methods. These approaches try to integrate concept learning as an intermediate step before making predictions, explaining the predictions in terms of human-understandable concepts. However, these methods require domain experts to extensively label concepts with relevant names and their ground-truth values. In response, we propose CAT, a novel interpretable Concept-bAsed Taylor additive model to simplify this process. CAT does not require domain experts to annotate concepts and their ground-truth values. Instead, it only requires users to simply categorize input features into broad groups, which can be easily accomplished through a quick metadata review. Specifically, CAT first embeds each group of input features into one-dimensional high-level concept representation, and then feeds the concept representations into a new white-box Taylor Neural Network (TaylorNet). The TaylorNet aims to learn the non-linear relationship between the inputs and outputs using polynomials. Evaluation results across multiple benchmarks demonstrate that CAT can outperform or compete with the baselines while reducing the need of extensive model parameters. Importantly, it can effectively explain model predictions through high-level concepts. Source code is available at github.com/vduong143/CAT-KDD-2024.

The task of finding Hierarchical Heavy Hitters (HHH) was introduced by Cormode et al. [12] as a generalisation of the heavy hitter problem. While finding HHH in data streams has been studied extensively, the question of releasing HHH when the underlying data is private remains unexplored. In this paper, we formalise and study the notion of differentially private HHH, in both the streaming and non-streaming setting. In the non-streaming setting, we show the surprising result that the relative error in estimating the count for any prefix is independent of the height of the hierarchy and the number of heavy hitters in the stream. Additionally, our algorithms also improve the error guarantees of Ghazi et al. [24] for the problem of counting over trees. Meanwhile, in the streaming setting, the main issue is that although the exact version of HHH has low global sensitivity (as counting queries are 1-sensitive), the approximation functions due to streaming have high global sensitivity, linear in the available space. Despite this obstacle, we show that the absolute error for estimating frequencies in the streaming setting is independent of the available space.

Reconfigurable intelligent surface (RIS) has drawn extensive attentions as a promising alternative for classical phased-array antennas at the massive multiple-input multiple-output (MIMO) transmitter, leading to cost-effective data transmission that is especially desirable at terahertz (THz) frequencies. In this article, we consider a multi-user MIMO (MU-MIMO) system equipped with a RIS-assisted transmitter: A RIS array is illuminated by unmodulated THz carriers through a feeding antenna, which is equally divided into a number of subarrays (SAs). Each activated SA serves one unique user equipment (UE) via directional beams. Then we develop a spectrum- and energy- efficient MU-MIMO scheme for THz communications by performing index modulation (IM) on the array-of-SA structure of the RIS, abbreviated as RIS-SA-IM. Specifically, the indices of the RIS-SAs allocated to different UEs, defined as SA allocation pattern (SAPs), are flexibly controlled by the information bits at each symbol period. Hence, aside from classical amplitude/phase modulation, additional energy-free bits (referred to asindex bits) can be conveyed implicitly by the chosen SAP at the transmitter, thus attaining superior enhancement on spectrum- and energy-efficiencies. Furthermore, we design a distributed mapping rule between the SAPs and index bits, which guarantees that the index information for each UE is exclusively determined by the index of its allocated RIS-SA. Hence, the proposed mapping rule can enable localized demodulation of the index bits without inter-UE data exchange. In particular, a general form of the distributed mapping rule is provided based on the binary-tree structure, which can be extended to arbitrary number of UEs and index bits. Additionally, the error performance of the proposed RIS-SA-IM is evaluated through pairwise error probability (PEP) calculations. Theoretical and simulation results demonstrate the superiority of our proposed RIS-SA-IM over its classical non-IM-aided counterpart.

The programmable metasurface (MTS), which can enhance the signal quality by flexibly manipulating the electromagnetic (EM) responses of reflected waves, has emerged as a promising technology for multiple-input multiple-output (MIMO) transmission due to its superior energy efficiency and cost-effective hardware implementations. In this paper, we consider a multi-user MIMO (MU-MIMO) system equipped with a MTS-based multi-feed transmitter, where an array of metallic elements is split into several subarrays (SAs), each irradiated by a unique radio-frequency (RF) feed. Additionally, each user is allocated with one unique RF-feed-SA pair for data services. Then we develop a novel index modulation (IM) scheme based on this array-of-SA (AoSA) structure of the MTS, named as MTS-SA -IM. More specifically, the allocation strategy of the SAs to different users, termed as SA allocation pattern (SAP), is flexibly controlled by the information bits. In other words, aside from classical amplitude/phase modulation bits, additional binary bits, referred to as index bits, can be embedded into the indices of the allocated SAs without extra power consumption, leading to enhancement of both spectrum and energy efficiencies. Furthermore, we propose a distributed mapping rule design between the index bits and SAPs, which guarantees that the index bits at each user are exclusively dependent on the index of its own allocated SA. By constructing a binary-tree structure, a general form of the proposed distributed mapping rule is generated recursively, which can be extended to the cases of arbitrary number of users. Simulation results demonstrate that the proposed MTS is capable of achieving desirable performance gain over its classical counterpart.

Reconfigurable intelligent surfaces (RISs), consisting of many low-cost elements that reflect the incident waves by an adjustable phase shift, have attracted sudden attention for their potential of reconfiguring the signal propagation environment and enhancing the performance of wireless networks. The passive nature of RISs is indeed beneficial, but the lack of radio frequency (RF) chains at the RIS has made channel estimation extremely challenging. We face this challenge by proposing a joint channel estimation and transmit precoding framework for RIS-aided multiple-input multiple-output (MIMO) systems. Specifically, the effective cascaded channel of the reflected transmitter-RIS-receiver link is decomposed into multiple subchannels, each of which corresponds to a single RIS element. Then our joint RIS-transmitter precoding model is formulated for the individual subchannels of each reflecting element. Finally, we develop a two-stage precoding design for successively determining the required phase shifts of each reflecting element of the RIS and the digital baseband precoder of the transmitter, only relying on the channel state information (CSI) of the subchannels. The performance of the proposed subchannel estimation and joint precoding method is evaluated by extensive simulations. Our numerical results show that the proposed designs provide an attractive solution to RIS-aided MIMO systems.

Reconfigurable intelligent surfaces (RISs), which can deliberately adjust the phase of incident waves, have shown enormous potentials to reconFigure the signal propagation for performance enhancement. In this paper, we investigate the RIS-aided offshore system to provide a cost-effective coverage of high-speed data service. The shipborne RIS is placed offshore to improve the signal quality at the vessels, and the coastal base station is equipped with low-cost reconfigurable reflect-arrays (RRAs), instead of the conventional costly fully digital antenna arrays (FDAAs), to reduce the hardware cost. In order to meet the rate requirements of diversified maritime activities, the effective sum rate (ESR) is studied by jointly optimizing the beamforming scheme and the service time allocated to each vessel. The optimal allocation scheme is derived, and an efficient fixed-point based alternating ascent method is developed to obtain a suboptimal solution to the non-convex beamforming problem. Numerical results show that the ESR is considerably improved with the aid of the RIS, and the proposed scheme using the hardware-efficient RRAs has only a slight performance loss, compared to its FDAA-based counterpart.

Predicting ambulance demand accurately at fine time and location scales is critical for ambulance fleet management and dynamic deployment. Large-scale datasets in this setting typically exhibit complex spatio-temporal dynamics and sparsity at high resolutions. We propose a predictive method using spatio-temporal kernel density estimation (stKDE) to address these challenges, and provide spatial density predictions for ambulance demand in Toronto, Canada as it varies over hourly intervals. Specifically, we weight the spatial kernel of each historical observation by its informativeness to the current predictive task. We construct spatio-temporal weight functions to incorporate various temporal and spatial patterns in ambulance demand, including location-specific seasonalities and short-term serial dependence. This allows us to draw out the most helpful historical data, and exploit spatio-temporal patterns in the data for accurate and fast predictions. We further provide efficient estimation and customizable prediction procedures. stKDE is easy to use and interpret by non-specialized personnel from the emergency medical service industry. It also has significantly higher statistical accuracy than the current industry practice, with a comparable amount of computational expense.