Relation Classification (RC) is a fundamental task in knowledge discovery. Prototypical networks are commonly used for few-shot RC tasks to tackle labeled data scarcity and long-tail relations, but they always overlook classification reliability and outlier features, potentially leading to sub-optimal similarity measurement. In a dynamic world with emerging novel semantic relations, incremental few-shot learning of relations gains attention, involving learning of both base and novel relations in new tasks while retaining base relation knowledge. However, this becomes challenging with increasing novel relations, requiring that models should reduce reliance on learning base relations for the new task. Therefore, this article explores Class-Incremental Few-Shot Relation Classification (CIFRC) using a Gaussian-Augmented Prototypical Network (GA-Proto). GA-Proto refines the similarity measurement by incorporating discrepancies between ideal and actual classifications via Gaussian Mixture Model and analyzing Gaussian outlier features of query instances. It also employs reliability learning and knowledge distillation to mitigate encoding space distortion and base relation forgetting by enhancing classification reliability and transferring base relation knowledge, respectively. Additionally, GA-Proto uses label smoothing to alleviate novel relation overfitting. Experimental results on three public datasets demonstrate that GA-Proto outperforms existing methods on CIFRC, achieving up to 19.64% improvement in accuracy. The datasets and source code for GA-Proto are released at https://github.com/GTZN2/GA-Proto .

The existing Discrete Event Simulators (DES) cannot meet the demands of modern networks for efficient, accurate, and flexible simulation. Recent m achine l earning models have demonstrated exceptional capabilities for e stimating n etwork p erformance (MLENP). However, the quality and quantity of available data greatly limit the accuracy and generalizability of ML models. After analyzing the data requirements of MLENP over the past decade and the shortcomings of existing DES, we propose a low-threshold and powerful n etwork p erformance data gen erator (GenNP), and generate a network performance dataset consisting of 10K samples. GenNP, with OMNeT＋＋ and INET at its simulation core, integrates the configuration generation layer, simulation transformation layer, result extraction layer, and result output layer, achieving massive random generation of simulation configurations (networks, traffic, routing protocols, faults) and multi-granularity extraction of network performance data (throughput, drop, delay, jitter, routing table). We validate the robust capabilities of GenNP through a series of simulation experiments across multi-granularity (spatial, temporal), diversity (traffic models, network load, fault types, routing protocols), and efficiency (parallelism).

Abstract Temporal Scoping of Facts is crucial for completing the temporal dimension of knowledge graphs. Current mainstream methods rely heavily on external resources for mining temporal information. However, the presence of noise in external resources, coupled with limitations in adaptively inferring non-continuous temporal dimensions with multiple temporal ranges, leads to low accuracy in predicting temporal ranges. To address these challenges, a model named JammyTS is proposed, which J oins an a ttention m echanism and a m emor y network for T emporal S coping of facts. Specifically, JammyTS leverages attention to adjust the distribution of weights dynamically in memory networks and builds attention capsule-based networks to reduce the impact of noise in external resources. Furthermore, two linear classifiers are separately trained to infer the end and beginning timestamps of facts for inference of non-continuous temporal ranges. Extensive experiments on three datasets show that JammyTS improves the accuracy by up to 12.29% compared to the state-of-the-art.

Abstract Named Entity Recognition (NER) in cybersecurity is crucial for mining information during cybersecurity incidents. Current methods rely on pre-trained models for rich semantic text embeddings, but the challenge of anisotropy may affect subsequent encoding quality. Additionally, existing models may struggle with noise detection. To address these issues, we propose JCLB, a novel model that J oins C ontrastive L earning and B elief rule base for NER in cybersecurity. JCLB utilizes contrastive learning to enhance similarity in the vector space between token sequence representations of entities in the same category. A Belief Rule Base (BRB) is developed using regexes to ensure accurate entity identification, particularly for fixed-format phrases lacking semantics. Moreover, a Distributed Constraint Covariance Matrix Adaptation Evolution Strategy (D-CMA-ES) algorithm is introduced for BRB parameter optimization. Experimental results demonstrate that JCLB, with the D-CMA-ES algorithm, significantly improves NER accuracy in cybersecurity.

With the booming development of fifth-generation network technology and Internet of Things, the number of end-user devices (EDs) and diverse applications is surging, resulting in massive data generated at the edge of networks. To process these data efficiently, the innovative mobile edge computing (MEC) framework has emerged to guarantee low latency and enable efficient computing close to the user traffic. Recently, federated learning (FL) has demonstrated its empirical success in edge computing due to its privacy-preserving advantages. Thus, it becomes a promising solution for analyzing and processing distributed data on EDs in various machine learning tasks, which are the major workloads in MEC. Unfortunately, EDs are typically powered by batteries with limited capacity, which brings challenges when performing energy-intensive FL tasks. To address these challenges, many strategies have been proposed to save energy in FL. Considering the absence of a survey that thoroughly summarizes and classifies these strategies, in this paper, we provide a comprehensive survey of recent advances in energy-efficient strategies for FL in MEC. Specifically, we first introduce the system model and energy consumption models in FL, in terms of computation and communication. Then we analyze the challenges regarding improving energy efficiency and summarize the energy-efficient strategies from three perspectives: learning-based, resource allocation, and client selection. We conduct a detailed analysis of these strategies, comparing their advantages and disadvantages. Additionally, we visually illustrate the impact of these strategies on the performance of FL by showcasing experimental results. Finally, several potential future research directions for energy-efficient FL are discussed.

The proliferation of various mobile devices with massive data and improving computing capacity have prompted the rise of edge artificial intelligence (Edge AI). Without revealing the raw data, federated learning (FL) becomes a promising distributed learning paradigm that caters to the above trend. Nevertheless, due to periodical communication for model aggregation, it would incur inevitable costs in terms of training latency and energy consumption, especially in multi-cell edge networks. Thus motivated, we study the joint edge aggregation and association problem to achieve the cost-efficient FL performance, where the model aggregation over multiple cells just happens at the network edge. After analyzing the NP-hardness with complex coupled variables, we transform it into a set function optimization problem and prove the objective function shows neither submodular nor supermodular property. By decomposing the complex objective function, we reconstruct a substitute function with the supermodularity and the bounded gap. On this basis, we design a two-stage search-based algorithm with theoretical performance guarantee. We further extend to the case of flexible bandwidth allocation and design the decoupled resource allocation algorithm with reduced computation size. Finally, extensive simulations and field experiments based on the testbed are conducted to validate both the effectiveness and near-optimality of our proposed solution.

IEEE INFOCOM 2023 - IEEE Conference on Computer Communications, New York City, NY, USA, 17-20 May 2023

Reliable data measurement is considered to be one of the critical ingredients for variant Internet of Things (IoT) applications. Gaining full knowledge of measurement data is becoming increasingly crucial to ensure a satisfactory user experience. However, data missing and corruption are inevitable in practical applications, which motivates us to study how to accurately recover the missing IoT measurement data in the presence of outliers. The data recovery problem can be formulated as a tensor completion (TC) problem. Existing TC methods are built on the assumption that the rank of the underlying tensor is fixed, which is not suitable for long data sequences in practice. Consequently, based on the characteristics of IoT streaming data, we assume that the data tensor lies in time-varying subspace, and an accurate estimate of the rank is a prerequisite for filling the missing entries and achieving robustness of the variations in both rank and noise. We built up an updatable framework based on dynamic CANDECOMP/PARAFAC (CP) decomposition. In addition, an efficient algorithm, called temporal multi-aspect streaming (T-MUST), is introduced to solve the optimization problem that originates in our developed model. It is worth noting that the proposed algorithm allows time-varying tensor rank and enables the rank changes could be detected and tracked automatically. Theoretical analysis indicates that T-MUST enjoys a geometric convergence rate. Numerical experiments conducted on various synthetic and real-world datasets empirically validate the superiority of the proposed T-MUST in both efficiency and effectiveness.

Tensor completion (TC) attempts to estimate the missing tensor entries from partial observations. Over the past decade, a tremendous amount of work has focused on this problem, which has given rise to a few excellent TC methods. Thus, a critical question is how to verify the optimality of TC models, especially when the underlying tensor is unknown, which is a common scenario in practice. The only existing method available-data validation-addresses the problem based on the validation set. While straightforward, this method may lead to a nonoptimal model, especially in scenarios with high missing ratios, and noise. In this work, we propose a self-validation method, which accounts for a metric named the identifiability measure, defined based on the generalization of isomeric conditions in the tensor case. It is noteworthy that the identifiability measure can verify the optimality of TC models without the use of any validation set. Extensive numerical experiments conducted on both synthetic, and real-world datasets empirically validate the superiority of the proposed method over the data validation approach.