From deep archival to real-time applications: Challenges and opportunities in DNA data storage.

Wildfire detection and localization in aerial imagery is critical for rapid response and damage mitigation. Autonomous aerial vehicles (AAVs) enable large area monitoring but face real-time processing challenges due to limited onboard computational and sensor resources. This work introduces a cross-modal knowledge distillation framework for edge-deployed AAVs. A teacher network trained only on thermal images transfers semantic and spatial representations to an optical image based student network when trained in an offline fashion using thermal and optical image pairs. During deployment, the student uses only optical images, thus reducing reliance on multi-sensor payloads while maintaining high detection accuracy. The student model incorporates dual classification heads: an image-level head for fire-free vs. fire-impacted scenes, and a patch-level head for flame vs. no-flame discrimination. This patch-level strategy provides effective fire localization while avoiding the computational overhead of segmentation, making it practical for resource-constrained deployment. Evaluated on aerial wildfire dataset, the framework achieves 90.97% patch-level accuracy, with false alarm and missed detection rates of 8.82% and 14.78%, respectively. The lightweight student model requires only 2.99 GFLOPS with inference time of 0.004s and generates patch-level probability heatmaps for fire region localization. Unlike conventional unimodal systems, this approach leverages thermal-to-optical knowledge transfer to deliver high accuracy, low latency, and precise localization under edge-computing constraints. The code and dataset will be released at https://github.com/medh132/cmkd .

Cloud gaming platforms lower the access barriers to graphics-intensive games by rendering computationally heavy game scenes on cloud GPU servers and streaming them back to players as real-time video, which in turn places significant demands on carrier networks to deliver these video streams with high throughput, low latency and minimal packet loss. To achieve decent user experience, cloud gaming platforms adapt streaming behaviors based on network conditions and allow users to adjust their graphics settings. Knowing the level of game streaming adaptability offered by various cloud gaming providers is helpful for network operators to effectively provision network resources for subscriber satisfaction, and for game development community to incentivize cloud gaming providers to better optimize their streaming techniques. Toward this objective, we develop a systematic framework to assess the adaptability of a cloud gaming platform in reducing network demand for lower client graphics settings; and in adjusting streaming quality under constrained network conditions for smooth gaming experience. Focusing on four popular platforms (NVIDIA GFN, Xbox, PlayStation and Amazon Luna), we begin by empirically profiling and comparing how they adapt game streaming characteristics to various levels of client graphics settings and network conditions. Building on the insights, we develop our systematic assessment framework, which provides quantitative scores for both fine-grained metrics by processing labeled traffic traces, as well as aggregated scores tailored to an assessor’s preference. We showcase our quantitative assessments of the four platforms.

The rapid proliferation of agentic AI systems has been dominated by Large Language Models (LLMs), but their substantial operational costs and high latency present significant barriers to widespread adoption. In response, the research community has increasingly turned to Small Language Models (SLMs), which offer a compelling combination of efficiency, task-specificity, and cost-effectiveness. This paper introduces the Adaptive Agentic Meta-Controller (AAMC), a deep reinforcement learning (RL) framework designed for intelligent SLM/LLM orchestration. The AAMC transforms the model selection problem into a principled, multi-objective optimization task, learning a dynamic policy that routes queries to the most appropriate model—preferring SLMs for routine tasks and escalating to LLMs only when necessary. Our framework features a Task Complexity Estimator (TCE) and an RL-based Router (RLR) that collaboratively balance the trade-offs between performance, cost, and latency. We conduct extensive experiments in a high-fidelity simulation environment, demonstrating that the AAMC achieves a task success rate comparable to an LLM-only approach while reducing operational costs by over 70% and significantly improving inference latency. We further introduce a comprehensive set of experiments on robustness, scalability, and fairness, including new ablation studies on the impact of the TCE and the sensitivity to user preferences, alongside a detailed complexity analysis and a discussion of real-world deployment. We further introduce a comprehensive set of experiments on robustness, scalability, and fairness, alongside a detailed complexity analysis and a real-world enterprise deployment case study. We also release code and experiments to support reproducibility.

Real-time sign language translation systems frequently suffer from high latency and bandwidth bottlenecks due to continuous video frame transmission and computationally heavy backend processing. This paper introduces Sign Recognition Model, an end-to-end American Sign Language (ASL) to spoken audio translation system designed for low-latency communication. To minimize redundant data transfer, Sign Recognition Model employs an edge-optimized frontend utilizing React.js and MediaPipe.js, which tracks hand visibility and selectively captures frames at 500ms intervals only when continuous signing is detected for over one second. These contextually rich frames are transmitted via a REST API to a Python FastAPI backend. The system utilizes a Long Short-Term Memory (2D-CNN) network trained on the processed Word-Level American Sign Language (WLASL) dataset, extracting both manual (hand signs) and non-manual (facial expressions) features to accurately decode sequences into text. To further reduce latency, the predicted text is queried against a Redis cache; if a cache miss occurs, the text is synthesized into natural-sounding audio using Coqui TTS or Suno AI’s Bark model, asynchronously cached, and streamed back to the client for automated playback. By distributing the computational load between client-side landmark detection and a cache-optimized backend, Sign Recognition Model provides a scalable, near real-time auditory communication bridge for the Deaf and Hard of Hearing community.

The increasing adoption of distributed systems has created a strong demand for secure and efficient data exchange, exposing the limitations of traditional centralized architectures. Conventional file-sharing systems rely on a single server for authentication and storage, which leads to issues such as single-point failure, identity exposure, and limited scalability. These systems often lack robust identity verification across distributed nodes, making them vulnerable to impersonation, interception, and unauthorized access. Additionally, their dependence on static credential mechanisms and inefficient communication models results in higher latency and reduced system performance. Such limitations, including centralized identity management, increased processing overhead, and restricted collaboration between nodes, highlight the need for a decentralized and lightweight secure framework. To overcome these challenges, a system is developed using a Django-based web platform integrated with a Peer-to-Peer (P2P) architecture and Elliptic Curve Cryptography (ECC). The proposed system utilizes Elliptic Curve Diffie–Hellman (ECDH) as a key agreement mechanism to securely establish a shared secret between peers. When a requested file is not available locally, the system initiates secure ECC-based authentication with another peer and retrieves the file without relying on a central authority, ensuring confidentiality, integrity, and anonymity during communication. The Django framework effectively manages user interactions, database validation, file processing, and performance visualization, while the use of ECDH enables strong security with smaller key sizes, resulting in faster execution and improved overall efficiency.

Real-time perception of dynamic visual scenes requires efficient extraction of spatiotemporal features. However, conventional image sensors fail to capture inter-pixel correlations, leading to redundant data transfer, high power consumption and latency. Here, we present a non-pixelated in-materia retinomorphic sensor (IMRS) that exploits the intrinsic spatiotemporal dynamics and correlated distributions of photocarriers for visual information processing. Built on a large-area graphene/silicon heterostructure, the IMRS integrates circumferentially arranged sampling electrodes that harness the lateral photovoltaic effect to convert incident optical patterns into spatial carrier distributions, which are further encoded as object-shape-dependent photovoltages. Mimicking the lateral inhibition of biological retinas, this sensor enables in-sensor spatiotemporal perception without image reconstruction. We demonstrate human motion recognition with over 98% accuracy while compressing raw visual data from 10,000 to 48 bytes, reducing postprocessing networks parameters by two orders of magnitude. These results establish spatiotemporal photocarrier dynamics in low-dimensional heterostructures as a computational primitive for energy-efficient, ultralow-latency processing of high-dimensional spatiotemporal information.

Remote collaboration systems based on physical environments face several critical challenges, including data-heavy virtual representations and high latencies during data acquisition, reconstruction, rendering, and transmission. Existing approaches often suffer from significant latency, making them unsuitable for real-time collaboration, rely on static scenes that limit interaction, and require multiple specialized hardware, restricting accessibility. To address these challenges, we present Collaborative Interactive Dynamic Environments for eXtended Reality (CIDER)—the first eXtended Reality (XR) platform to integrate Mixed Reality (MR) for co-located users and Virtual Reality (VR) for remote participants through a fully automated pipeline that replicates entire physical environments. CIDER dynamically transforms a user’s physical space into an interactive virtual environment, shareable with remote collaborators within seconds. It employs an efficient approach to represent, render, distribute, and synchronize virtual scenes, achieving interaction latencies of 0.22 seconds, about 10 times lower than comparable systems (2.4 seconds). We evaluate CIDER’s performance quantitatively with collaboration-oriented metrics in scenarios where participants are separated by up to 12,000 km. We also conducted a questionnaire-based user study with 17 participants to evaluate usability and overall user experience. Furthermore, CIDER allows collaborators to participate using a broad range of devices, including personal computers (via Unity emulators, functioning similarly to a MR/VR device), MR devices (e.g., HoloLens 2), and VR devices (e.g., Meta Quest 2 and 3), enhancing accessibility and usability for diverse user groups.

The rapid evolution of underwater communication technologies has become essential for applications such as marine exploration, environmental monitoring, and defense systems. Historically, underwater communication has relied primarily on acoustic methods due to their ability to travel long distances through water. However, with the increasing demand for high-speed and reliable data transmission, traditional systems face significant challenges. Acoustic communication, while effective over long ranges, suffers from low data rates, high latency, and susceptibility to noise and signal distortion. Similarly, radio frequency (RF) communication is highly inefficient underwater due to severe attenuation. These limitations highlight the need for an advanced communication approach capable of providing faster and more efficient data transfer in underwater environments. In this context, Light Fidelity (LiFi) technology emerges as a promising solution by utilizing visible light for data transmission. The proposed system integrates LiFi for high-speed, short-range underwater data communication and acoustic communication for long-range transmission, thereby combining the advantages of both technologies. The system employs LED-based transmitters and photodetector receivers for optical communication, along with acoustic modules for extended reach. This hybrid approach enhances data transmission efficiency, reduces latency, and improves reliability in dynamic underwater conditions. The significance of this work lies in its ability to overcome the inherent limitations of conventional systems, offering a scalable and efficient solution for nextgeneration underwater communication devices, with potential applications in oceanographic research, underwater robotics, and naval operations.

Clinician burnout caused by extensive electronic health record (EHR) documentation is a significant issue, frequently diminishing the time allocated for patient engagement. To solve this problem, we developed MediScribe AI, an ambient clinical intelligence system that uses speech recognition, speaker diarization, and clinical entity extraction to automate medical documentation. The system uses OpenAI Whisper, MedCAT, ClinicalBERT, and GPT-4o to generate structured SOAP notes and enable EHR updates in real time. In this study, we examined 29 recent publications (2024–2025) and found that ambient AI scribes can substantially decrease documentation time and cognitive load. However, errors in speech recognition and named-entity recognition (NER) still require human review. MediScribe AI aims to achieve high accuracy, low latency, and strong usability, making it suitable for healthcare settings where resources are limited.

We present an algorithm for efficient evaluation of Boys functions F0,…,Fkmax tailored to modern computing architectures, in particular graphical processing units, where maximum throughput is high and data movement is costly. The method combines rational minimax approximations with upward and downward recurrence relations. The non-negative real axis is partitioned into three regions, [0, ∞⟩ = A ∪ B ∪ C, where regions A and B are treated using rational minimax approximations and region C by an asymptotic approximation. This formulation avoids lookup tables and irregular memory access, making it well-suited for hardware with high maximum throughput and low latency. The rational minimax coefficients are generated using the rational Remez algorithm. For a target maximum absolute error of ɛtol = 5 × 10-14, the corresponding approximation regions and coefficients for Boys functions F0, …, F32 are provided in AppendixD.

ABSTRACT Underwater wireless sensor networks (UWSNs) have been used increasingly for critical tasks such as environment surveillance and underwater exploration. Nevertheless, their peculiar working environment, which entails their use of acoustic communication, high message latency, low bandwidth, mobility, stringent energy resources, and noisy communication channels, renders them susceptible to complex cyber threats including Sybil, Denial of Service (DoS), and traffic analysis attacks. This reality is a major drawback to the use of traditional intrusion detection systems used in other wireless communication networks. This manuscript addresses these issues through the presentation of an optimized cyberattack detection framework in UWSNs using an equivariant quantum convolutional neural network integrated with flamingo jellyfish search optimization (CD‐UWSN‐EQCNN‐FJSO). In this approach, network traffic data from the NSL‐KDD dataset are normalized using Bayesian boundary trend filtering (BBTF) to handle noise and uncertainty. Bitterling fish optimization (BFO) is then applied for feature selection, with further statistical feature extraction made by the double probability integral transform (DPIT). Here, the designed equivariant quantum convolutional neural network (EQCNN) is well leverage equivariance properties to perform robust detection under dynamic underwater network conditions, while the flamingo jellyfish search optimization (FJFO) approach dynamically fine‐tunes the weights within the network for improved detection accuracy and lower false alarm rates. The experimental results indicate that the proposed CD‐UWSNs‐EQCNN‐FJSO approach is able to provide much 27.15%, 26.09%, and 28.10% higher accuracy and 29.03%, 25.23%, and 29.1% higher recall capabilities, as well as much lower rates of 20.09%, 22.24%, and 20.01% lower false positive rate than other UWSN‐based cyber security solutions available in the existing methods.

The rise of connected devices since the industrial revolution has led to vast data generation and new digital challenges. A huge data from smart assets demanded scalable, private, and low-latency solutions. We propose a fog computing approach that brings analytics closer to devices. Our system enhances a standard machine-to-machine architecture using container-based orchestration for autonomy and peer-to-peer cyber-physical system communication. The focus is on smart factories and industrial Internet of Things (IIoT) applications. Recent progress on lightweight deep learning algorithms and fog computing permits multiple model inference tasks to run simultaneously on these resource-limited edge devices, so that we can collaboratively make one thing instead of getting good model quality in each single task. However, the high running latencies overall in multi-model inferences are a drawback for real-time applications. The proposed method introduces a hybrid partial swarm optimization-genetic algorithm scheduler that merges particle swarm optimization and genetic algorithm techniques to fine-tune task initiation times and minimize latency. By leveraging the strengths of both algorithms, it dynamically updates scheduling decisions for enhanced efficiency. This AI-driven model integrates IoT and digital twins to support adaptive, real-time optimization in smart manufacturing environments. Its innovation lies in balancing complex trade-offs across multiple objectives, delivering significant gains in agility and performance within the Industry 4.0 paradigm.

Abstract Nowadays, in the field of volcanic monitoring, as the number of monitoring stations, the related measured parameters and the complexity of the analyses performed on them increase, the traditionally used software architectures may prove inadequate. In this work we present a novel architecture to perform complex real-time analysis of multi-disciplinary data for volcanoes monitoring. The architecture is based on the idea of a set of independent workers exchanging messages on a distributed high performance message system. Workers are properly orchestrated in order to provide high throughputs and low latency times. A first platform implementing the architecture is presented with a real use case on Stromboli volcano monitoring where the Neolo system has been developed, and is described. Along this, it is described the adoption of the new platform also on the INGV-OV Surveillance Room. Used technologies and some workers that have been developed are also presented.

Underwater wireless sensor networks (UWSNs) are essential to the work of the navy, as they are used to monitor objects (surveillance), to monitor the environment (environmental monitoring), and to defend the tactics (tactical defense). They are however challenged by serious issues in deployment because of limitation of underwater communication by acoustic means such as high latency, low bandwidth, high rate of packet loss and extreme energy limitation. The conventional centralized approach of data processing cannot survive under these circumstances and a shift towards the decentralized intelligence is needed. This paper will present an Energy-Aware Clustered Federated Learning (CFL) framework, which is specific to UWSNs in naval systems. The approach suggested will arrange sensor nodes into logical cluster, where local models are being trained and aggregated at cluster heads and sent to a central unit. In order to extend the network lifetime, an energy-conscious participation scheme is used to make sure that only nodes that are energy-reliant participate in model training. Moreover, we propose a powerful median-based aggregation approach at the cluster level in order to overcome the impacts of underwater communications that are noisy and lossy. Simulations conducted under realistic conditions in the underwater environment prove that the proposed CFL architecture is much more accurate in models, can boost communication overhead, and increase the energy efficiency of the system as opposed to conventional federated learning tools. It is also demonstrated that the results are much more robust to packet loss and communication failures, confirming the relevance of the framework in autonomous underwater operations. This paper points out the potential transformations that federated learning can bring to allow the development of intelligent, resilient, and energy-efficient, underwater sensor networks, and create new opportunities in the future in the fields of naval and maritime applications in challenging underwater conditions. Keywords: Underwater Wireless Sensor Networks (UWSNs); Federated Learning; Energy Aware Systems; Clustered Aggregation; Naval Applications.

The demand for real-time streaming graph analysis has grown significantly, as hundreds of thousands of updates come every second. Monotonic graph algorithms such as Shortest Path are widely used in real-time analytics, but there are two bottlenecks that limit their performance, specially on planar graphs. One is massive redundant data accesses due to irregular state propagations and the other is high memory latency caused by irregular data accesses. We observe that existing systems mainly focus on general-purpose graph algorithms. If the properties of specific graph algorithms are exploited, the analysis performance can be further improved. Moreover, these systems typically tackle these two bottlenecks separately through either software or hardware mechanisms, but not both. However, both bottlenecks need to be addressed simultaneously in real scenarios such as road navigation. This paper proposes WSGraph, a software-hardware co-design framework for high-performance streaming graph processing. WSGraph tackles these two challenges by enforcing regularized processing orders and enabling precise data prefetching. Specifically, at the software level, WSGraph integrates a priority-based work scheduler with sliding-window bucket mapping scheme to regulate state propagations, thereby drastically reducing redundant data accesses. At the hardware level, WSGraph incorporates a lightweight in-core Proactive Data Engine (PDE). By exploiting intra-vertex access regularity, the PDE accurately prefetches relevant graph data to effectively hide the high latency of irregular memory accesses. Experimental results demonstrate that WSGraph achieves significant performance improvements over existing systems. Compared with the state-of-the-art software system KickStarter, WSGraph gains a 2.13 × speedup primarily by reducing graph data accesses by an average of 78.6%.

To address the difficulties in accurately capturing the characteristic changes of track irregularities in real - time and the limited ability to process complex mixed vibration signals, this study proposes an online Detection of Railway Track Irregularities via JADE-Based Blind Source Separation and MEMS Accelerometry. The system consists of a lower computer and an upper computer with ADXL345 three-axis acceleration sensor as the core. Real time track vibration signals are collected through optimized IIC bus protocol, and the blind source separation algorithm based on JADE is executed by STM32F103ZET6 microprocessor. By jointly diagonalizing the mixed vibration signal through a fourth-order cumulative matrix, the track roughness feature components in the mixed vibration signal are effectively decoupled, achieving accurate detection of railway track roughness. The detection results are converted into USB signals through RS-232 serial port and CH340G chip, and uploaded to the upper computer. The upper computer platform visualizes the type, location, and severity of track roughness faults. At the same time, a dual level power management and anti reverse protection are designed to ensure the reliability of the railway environment. To verify system performance, 8 monitoring points were set up on a 30 kilometer actual operating line, and multiple sets of vibration data were continuously collected at a sampling frequency of 10240 Hz at train speeds of 60-80 km/h. Establish the ground truth value of faults through high-precision track inspection vehicles and total station measurements, and compare it with HybridGAN method and data mining method. The experimental results show that this system can achieve an average positioning error of ≤ 1.8 mm, a fault type recognition accuracy of ≥ 96%, and an average detection time of ≤ 90 ms at a speed of 60 km/h. At a speed of 80 km/h, it still maintains an error of ≤ 2.2 mm and a recognition accuracy of ≥ 90%, with better performance than the two comparison methods. The upper computer of the system has the function of visualizing fault types, locations, and degrees, and integrates dual level power management and anti reverse protection, which is suitable for complex railway environments. This system provides a feasible solution for real-time monitoring of track status with high accuracy and low latency.

Drones have proven effective for acquiring aerial imagery, and when equipped with onboard analysis tools, they can automatically identify objects of interest. Neural-network methods for image analysis typically require large training datasets and substantial computational resources. By contrast, algorithmic techniques can detect objects using simple features, such as pixel colors, thereby reducing the need for extensive training and computational resources. Once trained, both types of system can analyze images in a short time. In our experiments, each approach has distinct strengths. The YOLO-based detector is more accurate for complex-shaped objects, such as trees, whereas the pixel-color approach performs better on sparser objects. This paper proposes YOLO-C3, a hybrid system designed for onboard drone image processing. By leveraging the strengths of both YOLO-based and pixel-based approaches, YOLO-C3 balances detection accuracy with estimation confidence. Trained on Mediterranean imagery dataset, the system is optimized for identifying natural objects, including citrus groves and trees. To assess the robustness of the image classifier, a K-fold cross-validation is performed. Compared to existing models, YOLO-C3 detects a wider range of natural objects with high accuracy and minimal latency, achieving a processing speed of 0.01 s per image. By performing object detection locally, drones can adapt their trajectories to support emergency response, helping to map safe corridors and locate buildings where people may be awaiting rescue after a natural disaster.

The rapid evolution of the Internet of Things (IoT) has enabled large-scale deployment of sensor networks for environmental monitoring in domains such as smart cities, agriculture, industrial automation, and disaster management. However, traditional cloud-centric IoT architectures face significant challenges, including high latency, bandwidth limitations, data privacy concerns, and unreliable connectivity. These limitations hinder real-time decision-making, which is critical for applications such as air quality monitoring, flood prediction, and wildfire detection. To address these challenges, Edge Artificial Intelligence (Edge AI) has emerged as a transformative paradigm that integrates AI capabilities directly at or near the data source. This research proposes an Edge AI-enabled IoT architecture designed to achieve low-latency and efficient smart environment monitoring. The architecture leverages distributed intelligence by deploying lightweight machine learning models on edge devices such as microcontrollers, gateways, and embedded AI processors. By performing real-time data processing and inference locally, the system significantly reduces dependency on cloud infrastructure, minimizes communication delays, and enhances system responsiveness. The proposed system adopts a multi-layer architecture consisting of the sensing layer, edge processing layer, communication layer, and cloud layer. Environmental data such as temperature, humidity, air quality index (AQI), and noise levels are captured using IoT sensors. Edge nodes process this data using optimized AI models (e.g., TinyML-based classifiers) to detect anomalies and generate alerts in real time. Only relevant or aggregated data is transmitted to the cloud for long-term storage, advanced analytics, and model updates. Experimental results demonstrate that the proposed architecture reduces latency by up to 60–80% compared to traditional cloud-based systems, while maintaining high prediction accuracy (>92%). Furthermore, bandwidth consumption is reduced significantly due to localized processing. The system also enhances data privacy by minimizing the transmission of sensitive raw data. The integration of communication protocols such as MQTT and LoRaWAN ensures efficient data transmission in resource-constrained environments. This study contributes to the advancement of intelligent IoT systems by providing a scalable, energy-efficient, and low-latency architecture for smart environment monitoring. The findings indicate that Edge AI is a viable solution for next-generation IoT applications requiring real-time analytics and adaptive decision-making.

Indoor localization in complex architectural environments remains a significant challenge due to the inability of Global Positioning System (GPS) signals to penetrate concrete structures. This paper introduces Sanjaya, an innovative and cost-effective Indoor Positioning System (IPS) designed for real-time student monitoring in campus environments. The proposed system utilizes WiFi Fingerprinting methodology, specifically employing a Deterministic K-Nearest Neighbour (k-NN) algorithm based on Euclidean Distance metrics to map Received Signal Strength Indicator (RSSI) values to specific indoor locations. Unlike traditional hardware-intensive solutions like RFID or BLE, Sanjaya leverages existing campus WiFi infrastructure, significantly reducing deployment costs and maintenance requirements. The system architecture integrates a NodeMCU (ESP8266) edge device for signal acquisition with a Flask-based Python backend for algorithmic processing. Experimental results demonstrate that the system achieves high room-level accuracy and low latency, providing a scalable and privacy-conscious solution for enhancing safety and situational awareness in educational institutions.