Graphs play a crucial role in various research fields, including network science, machine learning, and data analysis. Algorithms based on random walks have become popular for information diffusion, target node research, and exploration on graphs. Random walk algorithms can be classified into local and non-local random walks. Local random walks enable transition between adjacent nodes based on predefined probabilities, while non-local random walks enable transitions that extend beyond immediate neighbors. While non-local random walks enable faster graph exploration by reaching beyond immediate neighbors, these methods typically require the agent to have full knowledge of the entire graph and the ability to freely move to any non-adjacent node, which makes them difficult to implement in large-scale networks. The aim of this study is to develop a local random walk that improves graph exploration efficiency using only limited non-local information, without requiring the agent to have knowledge of the global graph structure. Specifically, we propose a random walk called Sprint Walk, in which certain nodes in the graph record the shortest paths to a small number of anchor nodes, thereby enabling non-local-like transition behavior by the agent. Through simulation experiments, we demonstrate that Sprint Walk significantly reduces search and exploration time compared to classical local random walk algorithms, including the Simple Random Walk (SRW), Biased Random Walk (BiasedRW), Non-Backtracking Random Walk (NBRW), and Self-Avoiding Random Walk (SARW) across nine graph types.

Random walk-based algorithms are a key technique in many engineering applications, such as graph exploration, information diffusion, social network analysis, recommendation systems, machine learning, and biological network modeling, due to their adaptability, scalability, and simplicity. Although previous research has demonstrated that random walks exhibit vulnerability to certain link-rewiring attacks, a broader range of adversarial attack strategies remains largely unexplored. In this paper, we systematically investigate how diverse link deletion methods — bridge removal, hub removal, and exit blocking — combined with three link addition methods — triangle, loopback, and cluster trapping — impact the target node search (hitting time) and graph exploration (cover time) of the most standard types of random walk algorithms, such as Simple Random Walk, Degree-based Random Walk, and Memory-based Random Walk, across four types of graphs (Erdos and Rényi, Barabási-Albert (BA), voronoi, and regular) through simulation experiments. We also propose a mitigation strategy called the link reinforcement strategy, which strengthens the resilience of random walks against the aforementioned nine adversarial attacks. This paper highlights the weaknesses of random walk algorithms against adversarial attacks and presents an effective method to mitigate performance degradation caused by these attacks.

In distributed learning, the parameters of a regression or classification model are estimated from training data ob-served at each node distributed in a network without aggregating the data at one place. In particular, approximating a kernel function using Random Fourier Features (RFF) enables large-scale distributed learning. In the literature, Random Fourier Features distributed online kernel-based learning (RFF-DOKL) has been proposed; however, the impact of the network topology on the performance of RFF - DO KL has not been well understood. In this paper, we investigate the impact of the network topology among learning nodes on the performance of RFF - DO KL. Furthermore, we clarify the conditions and underlying factors that influence the performance. Specifically, we experimentally examine the convergence speed, model accuracy, and total communication cost (i.e., total number of message exchanges among learning nodes) in different types of network topologies with the same number of learning nodes.

Graphs serve as powerful representations for various real-world systems such as social networks, biological networks, and communication networks. Random walk algorithms have gained popularity for graph-based data analysis and processing, finding applications across various domains. Understanding and enhancing these algorithms is crucial for ensuring high-quality protocols, controls, and services in large-scale communication networks. While conventional random walks typically rely on local information, there is potential to improve node search efficiency by incorporating information beyond the local context. Concurrently, there is growing interest in machine learning techniques that represent data as graphs rather than vectors, known as graph and node embedding algorithms. This paper investigates whether leveraging node embedding vectors generated by such techniques can enhance the efficiency and effectiveness of random walks on a graph. To address this, we propose EmbedRW (Embedded Random Walk), which integrates node embedding techniques with random walk design. Through simulation experiments, we demonstrate that utilizing node embeddings can significantly reduce the search time for the target node across a wide range of graphs.

Random walks are being used for target node search and graph exploration in various fields, including communications and social networking. However, frequent revisits of the random walk agent to the same node degrade the efficiency of node search and graph exploration. To address this issue and improve the efficiency of node search and graph exploration, a variety of random walk-based algorithms with history, such as self-avoiding random walk (SARW) and k-history random walk (k-History), have been developed. Although these approaches accelerate node search and graph exploration with a random walk agent, they require the agent to have a non-negligible amount of memory space to store many nodes on large-scale graphs. To address this issue, it is essential to clarify efficient memory management strategies for random walk agents. In this paper, we propose novel random walk-based algorithms called BloomWalk and CuckooWalk, in which an agent performs history-based random walks on a graph using a probabilistic data structure to record previously visited nodes in memory efficiently. Experimental results demonstrate that Bloom Walk and CuckooWalk enable efficient node search on unknown graphs even with the very limited memory capacity.

Random walk-based algorithms are frequently utilized to target node search and graph exploration in unknown graph structures. Unlike deterministic algorithms such as breadth-first search and depth-first search, target node search and graph exploration with random walk algorithms are expected to exhibit robustness against adversarial attacks because of their probabilistic nature. The characteristics of random walks when adversaries change the topology of the graph, known as adversarial attacks on random walks, have been just recently received attention. These attacks have been shown to significantly degrade the efficiency of target node search and graph exploration with random walks. However, questions regarding how robust random walks are against more realistic attacks persist. In this paper, we investigate adversarial attacks in the form of rewiring a limited number of links during target node search and graph exploration, particularly in scenarios where mobile agents employ random walk algorithms. The goal is to quantitatively determine how robust or vulnerable the conventional random walk algorithms are against link rewiring attacks. We consider three types of link rewiring attacks (centrality method, clustering method, and starting node method) and evaluate how they affect target node search (hitting time) and graph exploration time (cover time) of five major random walk algorithms on a graph - Simple Random Walk (SRW), Non-Backtracking Random Walk (NBRW), k-History random walk (k-History), Biased Random Walk (BiasedRW), and Vicinity Avoidance Random Walk (VARW) - in five graphs through simulation experiments.

MultiPath TCP (MP-TCP) allows the use of multiple paths between two end hosts for a single data transmission, extending the capabilities of SinglePath Transmission Control Protocol (SP-TCP). AIMD congestion control algorithm can be implemented on each subpath of the MP-TCP sender. However, the characteristics of the AIMD-type window flow control depend on the control parameters (α, β). There have been some guidelines proposed to select proper control parameters (α, β) that can achieve fair bandwidth sharing between SP-TCP and MP-TCP senders using existing MP-TCP congestion algorithms. However, current guidelines do not offer a solution that can simultaneously increase the throughput of an MP-TCP sender and ensure fairness between MP-TCP and SP-TCP senders. To address this issue, this study proposes a control parameter setting for an MP-TCP sender that can maximize the sender’s throughput and ensure fairness between SP-TCP and MP-TCP senders. We derive a fluid model of a network that includes SP-TCP and MP-TCP senders, describing the relationship between control parameters, the aggregate throughput of an MP-TCP sender, and the packet loss rate of a router.

Information-Centric Networking (ICN) has gained attention as one of the next-generation internet architectures that focuses on the data being transmitted rather than the hosts transmitting it. Due to the differences between ICN and TCP/IP networks, it is not possible to evaluate the performance of ICN using network simulators designed for TCP/IP. A number of studies have been conducted to develop ICN network simulators. However, further acceleration of ICN network simulators is expected to enable large-scale ICN network performance evaluation. In this paper, we analyze the performance bottleneck of the flow-level ICN simulator called FICNSIM (Fluid-based ICNSIMulator) by profiling its performance using the Julia language source code. Specifically, we identify the processing that is causing the performance bottleneck of FICNSIM and investigate the scalability of FICNSIM with respect to network scale.

This paper addresses a research question: - how accurately can a TCP throughput prediction model be constructed while preserving the privacy of a large number of Internet users? In the field of communication networks, accurate performance prediction of TCP flows is crucial for realizing high-quality services. In recent years, machine learning techniques have advanced and approaches for TCP throughput prediction based on centralized machine learning have emerged. However, approaches for TCP throughput prediction lack the privacy protection of Internet users and struggle to cope with a large amount of training data. Federated Learning (FL) is a novel decentralized machine learning paradigm that was introduced in 2017, allowing for multiple learning clients to collaboratively train the parameters of the global model. In this paper, we propose the Federated Learning-based PERFormance predictor (FL-PERF) of TCP flows, which builds a global TCP throughput prediction model using FL with multiple learning clients in a privacy-preserving manner. Through experiments, we investigate the accuracy of the TCP throughput prediction model obtained with FL-PERF through experiments and then discuss its privacy and scalability.