International Conference on Computing, Networking, and Communications : [proceedings]

Accurately and continuously assessing cyber risk remains a core challenge in modern network defense. Conventional Cyber Threat Intelligence (CTI) products rely heavily on observed Indicators of Compromise (IoCs), which provide only partial or lagging visibility into malicious activity. As a result, defenders often lack persistent, data-driven measures of how threat exposure evolves across network entities such as Autonomous Systems (ASes). This paper introduces a probabilistic framework for estimating the security risk of ASes using time-varying latent risk scores derived from network threat intelligence. The method integrates heterogeneous threat signals, including IP reputation measures, observed threat events, and associated severity indicators, within a Bayesian Gaussian Random Walk model that infers a latent risk trajectory for each AS. Evaluation using both synthetic and real-world threat datasets demonstrates that the framework captures meaningful temporal trends, quantifies uncertainty during periods of sparse observation, and provides a continuous view of AS-level threat exposure.

In this paper, we propose a performance metric in order to evaluate the performance of deadline-aware scheduling methods.By using linear programming to find the ideal off-line scheduling and evaluating the performance of real-time scheduling methods based on ideal scheduling, it is possible to clearly identify the differences from the ideal scheduling.This paper proposes a performance metric that focuses on four evaluation metrics and formulates liner programming to determine optimal off-line scheduling for a baseline for evaluating online scheduling algorithm. It also examines optimal scheduling for both known and unknown job occurrence patterns. Additionally, we report the performance evaluation of EDF algorithm, a representative deadline-aware scheduling method, using this metric.

Information-Centric Networking (ICN) has attracted significant attention as a network architecture optimized for efficient and scalable content dissemination. Moving beyond the conventional paradigm of name-based content retrieval, ICN can also support communication patterns associated with services and in-network computation. This paper aims to strengthen ICN by leveraging its core principles and introducing complementary mechanisms to improve efficiency, reliability, and performance within IP-based infrastructures. We present ICNx, an extension of the baseline ICN architecture that incorporates functional capabilities supporting adaptive and coordinated communication for both content delivery and function invocation. The study includes the architectural design and a prototype implementation on the open-source Cefore platform. In addition, a large-scale research initiative under Japan’s JST Moonshot R&D program is discussed as an illustrative use case motivating the proposed design. The results demonstrate the practical feasibility of the approach and highlight the evolving role of ICN, not merely as a content-delivery mechanism, but as a foundational communication substrate for scalable and resilient applications.

This work provides the first unified comparison of DQN, DDPG, PPO, and SAC for 5G NR uplink and downlink modulation control, revealing how core RL design choices uniquely impact physical-layer Quality of Service (QoS). Deep Reinforcement Learning (DRL) algorithms optimize QoS by dynamically switching modulations in 5G New Radio (NR) systems. Evaluations span both Physical Downlink Shared Channel (PDSCH) and Uplink Shared Channel (PUSCH). Metrics include Bit Error Rate (BER), throughput, latency, and path loss. Results show DDPG consistently outperforms others, offering insights into the role of policy entropy and exploration in wireless control systems.

This paper presents a novel framework for generating realistic video footage at novel viewpoints from 3D Gaussian splats using text prompts. Current approaches face two key challenges: incorporating high-dimensional semantic information into dense Gaussian representations without costly annotations and efficiently passing representations to a language model to generate smooth, aesthetically pleasing camera trajectories. To address these, we propose a two-component pipeline. First, our Sparse Spatial-Semantic (SSS) attribute Augmentor (SAug) enriches Gaussian splats with spatial-semantic features learned from multi-view images alone, along with our sparsity control technique maintaining compact representations despite millions of splats. Second, our LLM-based Motion Generator (L-MR) leverages Large Language Models to produce smooth 6D camera pose trajectories directly from text prompts and compact object-based representations using SAug outputs. Together, these components enable text-driven generation of realistic novel-view videos, with SAug providing semantically-aware scene representations and L-MR translating natural language into cinematic camera motions. Experiments demonstrate that our method generates semantically and spatially more accurate trajectories while keeping the memory and computation footprint of the rendering pipeline low.