Edge Computing Unlocked: From Secure AI to Agile Robotics and Sustainable Infrastructure

Latest 50 papers on edge computing: Nov. 2, 2025

Edge computing is rapidly transforming the landscape of AI/ML, bringing computational power closer to data sources and enabling real-time intelligence in diverse environments. This paradigm shift addresses critical challenges like latency, bandwidth limitations, and privacy concerns inherent in traditional cloud-centric models. Recent research breakthroughs are pushing the boundaries of what’s possible at the edge, spanning from robust resource management to novel hardware designs and secure distributed learning.

The Big Ideas & Core Innovations

At the heart of these advancements lies a common thread: optimizing performance under constrained resources while enhancing security and adaptability. Several papers delve into intelligent resource allocation and task offloading. For instance, “Strategic Task Offloading for Delay-Sensitive IoT Applications: A Game-Theory-Based Demand-Supply Mechanism with Participation Incentives” by Author One and Author Two from University of Technology, proposes a game-theoretic framework to optimize task offloading in delay-sensitive IoT applications, showing that incentivized participation significantly improves efficiency and reliability. Similarly, “Joint Computing Resource Allocation and Task Offloading in Vehicular Fog Computing Systems Under Asymmetric Information” by Dong In Kim from Sungkyunkwan University, introduces the JCRATOA algorithm, which tackles asymmetric information in vehicular fog computing to balance computational efficiency and fairness, outperforming existing methods.

Focusing on aerial and integrated systems, the paper “Cost Minimization for Space-Air-Ground Integrated Multi-Access Edge Computing Systems” by Weihong Qin et al. from Jilin University, introduces the MADDPG-COCG algorithm, a multi-agent deep reinforcement learning approach combined with convex optimization and coalitional game theory, to minimize costs in complex Space-Air-Ground Integrated Networks (SAGIN-MEC). Another significant contribution is “AirFed: Federated Graph-Enhanced Multi-Agent Reinforcement Learning for Multi-UAV Cooperative Mobile Edge Computing” by Zhiyu Wang et al. from The University of Melbourne, which integrates multi-agent reinforcement learning with federated learning to enable efficient UAV-based edge computing, using dual-layer Graph Attention Networks (GATs) for dynamic decision-making.

Security and privacy are paramount. “Confidential FRIT via Homomorphic Encryption” by Haruki Hoshinoa et al. from The University of Electro-Communications, introduces confidential FRIT, a framework for secure data-driven gain tuning in control systems using homomorphic encryption. This method performs exact matrix inversion on encrypted data, a significant leap from prior iterative approximations. On the other hand, “Attack on a PUF-based Secure Binary Neural Network” by Bijeet Kumar from IIT Kanpur, highlights vulnerabilities in secure Binary Neural Networks (BNNs) protected by Physical Unclonable Functions (PUFs), emphasizing the need for more robust key management.

Hardware and model efficiency are also major themes. “Unsupervised local learning based on voltage-dependent synaptic plasticity for resistive and ferroelectric synapses” by Fabien Alibart from Université de Sherbrooke, presents Voltage Dependent Synaptic Plasticity (VDSP) for energy-efficient online learning in neuromorphic systems using memristors. For large language models, Yingshi Chen’s “EOE: Evolutionary Optimization of Experts for Training Language Models” from University of California, Berkeley, introduces an evolutionary optimization framework that reduces LLM size and memory requirements, making them more feasible for edge deployment.

Under the Hood: Models, Datasets, & Benchmarks

These innovations are often built upon or introduce novel computational models, datasets, and benchmarks:

• JCRATOA Algorithm: Proposed in “Joint Computing Resource Allocation and Task Offloading in Vehicular Fog Computing Systems Under Asymmetric Information” (https://arxiv.org/pdf/2510.26256), this algorithm optimizes resource allocation and task offloading in vehicular fog computing, outperforming existing solutions in delay, throughput, and fairness.
• Confidential FRIT with ElGamal and CKKS: Featured in “Confidential FRIT via Homomorphic Encryption” (URL not provided, likely arXiv), this method uses homomorphic encryption schemes like ElGamal and CKKS for secure computation, offering comparable performance to conventional methods.
• VDSP (Voltage Dependent Synaptic Plasticity): Introduced in “Unsupervised local learning based on voltage-dependent synaptic plasticity for resistive and ferroelectric synapses” (https://arxiv.org/pdf/2510.25787), this learning rule enables energy-efficient online learning in neuromorphic systems with memristive devices.
• Bayes-Split-Edge Framework: Presented by Fatemeh Zahra Safaeipour et al. from the University of Kansas in “Bayes-Split-Edge: Bayesian Optimization for Constrained Collaborative Inference in Wireless Edge Systems” (https://arxiv.org/pdf/2510.23503), this framework leverages Bayesian optimization for split inference in wireless edge systems, tested with VGG19 on ImageNet-Mini and mMobile wireless traces.
• AirFed with Dual-Layer GATs: From “AirFed: Federated Graph-Enhanced Multi-Agent Reinforcement Learning for Multi-UAV Cooperative Mobile Edge Computing” (https://arxiv.org/pdf/2510.23053), this system utilizes dual-layer Graph Attention Networks (GATs) with GRU for spatial-temporal dependency modeling in UAV-based edge computing.
• TinyissimoYOLO: Highlighted in “Ultra-Efficient On-Device Object Detection on AI-Integrated Smart Glasses with TinyissimoYOLO” (https://arxiv.org/pdf/2311.01057), this family of sub-million parameter YOLO architectures supports up to 80 classes for real-time, low-power object detection on smart glasses. An open-source implementation is available via https://github.com/RangiLyu/nanodet.
• RowDetr: Introduced in “RowDetr: End-to-End Crop Row Detection Using Polynomials” (https://arxiv.org/pdf/2412.10525) by Rahul Harsha Cheppally and Ajay Sharda from Kansas State University, this transformer-based neural network uses polynomial representation for robust crop row detection, achieving real-time inference on NVIDIA Jetson Orin AGX. Code: https://github.com/r4hul77/RowDetr-v2.
• Neural Operators for TBI: “Real-Time Brain Biomechanics Prediction with Neural Operators: Toward Clinically Deployable Traumatic Brain Injury Models” (https://arxiv.org/pdf/2510.03248) by Anusha Agarwal et al. from Johns Hopkins, benchmarks Fourier Neural Operators (FNO), Factorized FNO, Multi-Grid FNO, and DeepONet for real-time brain biomechanics prediction, with code at https://github.com/Centrum-IntelliPhysics/Neural-Operator-for-Traumatic-Brain-Injury.
• MECKD Framework: Developed by BoneZhou et al. in “MECKD: Deep Learning-Based Fall Detection in Multilayer Mobile Edge Computing With Knowledge Distillation” (https://arxiv.org/pdf/2510.03601), this framework integrates knowledge distillation for efficient fall detection on edge devices. Code: https://github.com/BoneZhou/MECKD.

Impact & The Road Ahead

These recent breakthroughs underscore the incredible potential of edge computing to revolutionize various industries. From enabling agile robotics to enhancing cybersecurity and making healthcare more accessible, the impact is far-reaching. The work on UAV-enabled MEC systems, like “Energy-Efficient UAV-Enabled MEC Systems: NOMA, FDMA, or TDMA Offloading?” (https://arxiv.org/pdf/2510.22306), points towards more robust and sustainable aerial networks, while “Radiance Field Delta Video Compression in Edge-Enabled Vehicular Metaverse” (https://arxiv.org/pdf/2411.11857) paves the way for immersive vehicular metaverse experiences with reduced latency.

The emphasis on lightweight and energy-efficient AI, as seen in “ECORE: Energy-Conscious Optimized Routing for Deep Learning Models at the Edge” (https://arxiv.org/pdf/2507.06011) and “Vacuum Spiker: A Spiking Neural Network-Based Model for Efficient Anomaly Detection in Time Series” (https://arxiv.org/pdf/2510.06910), means AI can be deployed in previously inaccessible, resource-constrained environments. “Sequencing on Silicon: AI SoC Design for Mobile Genomics at the Edge” (https://arxiv.org/pdf/2510.09339) truly democratizes genomic analysis, bringing advanced medical capabilities to remote regions.

Looking ahead, the integration of multi-agent reinforcement learning, game theory, and federated learning will continue to drive dynamic resource management and secure collaboration across distributed edge systems, as discussed in papers like “Chronicles of Jockeying in Queueing Systems” (https://arxiv.org/pdf/2402.11061) and “A Survey on Scheduling Techniques in the Edge Cloud: Issues, Challenges and Future Directions” (https://arxiv.org/pdf/2202.07799). The path is clear: edge computing is not just an optimization; it’s a fundamental shift enabling intelligent, adaptive, and sustainable AI for a connected future.