July 19 ~ 20, 2025, Toronto, Canada
Roy Hisar Martahan Simanungkalit, Sapto Jumono, Muhammad Fachruddin Arrozi Adhikara, Agus Munandar & Jaka Suharna
Department of Economics and Business, Esa Unggul University, Jakarta, Indonesia
Green Digital Finance (GDF) combines digital technology with green finance goals, leveraging technologies such as blockchain and artificial intelligence to support sustainable investment and financial practices. This article presents a systematic review of the literature on GDF, focusing on recent developments, challenges, and opportunities. The research method uses Systematic Literature Review (SLR) with PRISMA guidance, including identification, screening, eligibility, and inclusion of articles from various academic databases. The results of the SLR show a significant increase in research on GDF over the past five years, with the implementation of digital technologies such as Blockchain, big data, fintech, and AI being the most frequently used. The study also highlights challenges such as lack of adequate regulation, resistance to technological change, and security concerns. In conclusion, the GDF has great potential to improve ecological sustainability, but it requires the establishment of an appropriate regulatory framework, comprehensive education initiatives, and multi-sectoral collaboration.
Green Digital Finance (GDF), Digital Technology, Green Finance, Blockchain, Big Data, Artificial Intelligence (AI), Green Investment, Environmental Sustainability, Systematic Review of the Literature (SLR)
Zhen Zhang, Bradley Department of Electrical and Computer Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA
This study proposes a behavior-specific filtering method to improve behavior classification accuracy in Precision Livestock Farming (PLF). While traditional filtering methods, such as wavelet denoising, achieved an accuracy of 91.58%, they apply uniform processing to all behaviors. In contrast, the proposed behavior-specific filtering method combines Wavelet Denoising with a Low Pass Filter, tailored to active and inactive pig behaviors, and achieved a peak accuracy of 94.73%. These results highlight the effectiveness of behavior-specific filtering in enhancing animal behavior monitoring, supporting better health management and farm efficiency
Meri¸c Demir¨ors, Ahmet Murat Ozbayo˘glu, and Toygar Akg¨un ¨, TOBB University of Economics and Technology, Ankara, Turkey
The proliferation of 5G technologies and the vast deployment of Internet of Things (IoT) devices have heightened the demand for optimal spectrum utilization, necessitating robust spectrum management strategies. In this context, an efficient energy detector employing wideband spectrum sensing within a 5G environment is essential for identifying underutilized frequency bands suitable for cognitive radio applications across multiple sub-bands. While cooperative spectrum sensing (CSS) can enhance the detection capabilities of energy detectors amidst noise uncertainty, its performance often deteriorates under low signal-to-noise ratio (SNR) conditions. This study proposes an improved CSS framework that combines Maximal Ratio Combining (MRC) with the K-out-of-N fusion rule to address noise uncertainty in a complex Gaussian environment across multiple sub-bands in cooperative wideband spectrum sensing. Comparative performance analysis confirms that this integrated approach enhances detection probability and maintains a low false alarm rate across various low SNR scenarios, significantly outperforming traditional cooperative and non-cooperative wideband spectrum sensing methods. These results highlight the potential for advancing cognitive radio technologies by optimizing detection algorithms to improve performance under challenging conditions.
Signal-Noise Ratio, Maximal Ratio Combining, Wideband Spectrum Sensing, Energy Detection, K-out-of-N fusion rule
Blessing C. Dike andCajetan M. Akujuobi, Center of Excellence for Communication Systems Technology Research, ECE Dept. Prairie View A&M University Prairie View, Texas, USA
Developments in the field of generative-AI have made it extremely difficult to distinguish artificially generated content from real content. As a result, their reliable detection has become more important. The topic of this research is detecting speeches that are generated by future generative-AI models in unknown languages. It focuses on answering ”With what information does a model distinguish fake audio from real audio, does it learn how spoken languages sound, or does it learn a specific trait of generated speech waves?” Multiple models are trained on various datasets to detect synthetic audio signals generated by generative-AI models. After multiple sessions of trains and tests, the best test accuracy scores for different test sessions are 94.92% for known language from unknown generative-AI model, 98. 44% for an unknown language from known generative-AI model, and 95. 18% for an unknown language from unknown generative-AI model.
CNN, Bispectrum
Adrian E. Conway, Assured Networking Solutions, Weston, MA 02493, U.S.A
A physical unclonable function (PUF)-based method is presented for continuously authenticating the physical hardware provenance of data that a sensor streams over time to a receiving device. In contrast to existing PUF-based authenticated remote sensing techniques, the method does not make use of any PUF challenge-response pair (CRP) databases or PUF models. The new method, that we call a Sensor Ratchet, is based on the hitherto developed PUF-based CRP Ratchet protocol for continuously mutually authenticating a pair of devices over time. As such, the Sensor Ratchet inherits the lightweight computational requirements and secure properties of the CRP Ratchet. Three variations of the Sensor Ratchet are developed: a simplex form of authenticated data transfer in which a receiver is the initiator, a simplex form in which the sensor is the initiator, and a half-duplex form that additionally transmits physically authenticated information - such as control signals - to a sensor.
Sensor data physical provenance, physical authentication, ratchet protocol, physical unclonable function, physical unclonable protocol
Krishnageetha Karuppasamy, Abinash Borah, Anirudh Paranjothi, and Johnson P Thomas
The Internet of Vehicles (IoV) offers various services for road safety and user comfort. However, they face security vulnerabilities such as false data injections which need to be mitigated for public safety. The security solutions for IoV should have minimal processing delay and be scalable to deal with the large-scale IoV. While classical machine learning techniques have been adopted for malicious node detection in IoV, these solutions face computational challenges and scalability limitations. To deal with these challenges, in this paper, we propose a novel quantum-based MaxCut graph detection mechanism for identifying malicious nodes transmitting false messages in IoV. As validated by the performance evaluation results, the proposed quantum-based detection approach offers significantly lower data processing delay compared to the classical approach, especially as the data size increases.
Internet of vehicles, security, false message detection, quantum
Mohammad Shadeed1 and Majdi Owda2, 1Department of Computer Engineering, University of Fairfax, VA, USA, 2Department of Computer Engineering, Arab American University, West Bank, Palestine
Timestamps have proven to be an important and appropriate source to some extent as a guide for investigators in discovering computer crimes. that computer crimes can be detected in several ways. they criminals always try to hide their gender, place, and date of the crime they committed, previous research focused on the importance of timestamp and in this research, we introduced new elements ($Logfile\ $MFT\$MFTMirr\ Windows event logs) that contain artifacts about timestamp in NTFS file system and provides valuable information during investigations to detect anomalies in NTFS file system timestamp, which is very important, as it can be used as strong proof of a computer crime, This paper deals with the reliability check of artifacts used to detect anomalies in the NTFS file system timestamp and the ability of these artifacts to keep records for users.
NTFS filesystem, Anomaly Detection, $logfile, timestamp, $MFT.
Shah Mehmood Wagan, Xinli Zhang and Sidra Sidra, Business School, Sichuan University, Chengdu, China
The study focuses on the innovations by artificial intelligence in the workplace that are affecting the productivity of the staff and the overall performance of businesses. It attempts to uncover the mechanism behind technologys technological impact on business operations and labor productivity. A quantitative research technique was used in this study with SmartPLS. It is found in a study that out of 350 small and medium-sized company samples, the first two had the highest adoption rates. The study specifies that perceived value, as well as the ease of use, has a major effect on the adoption of AI solutions. Technology development is one such method, through which increasing the level of work of people in the business raises the companys productivity. The positive experience increases business performance, therefore accuracy of business management to understand the employees of the business and customer satisfaction are positively related these points have been illustrated in this paper. Besides fact that study is mainly based on self-reporting data it may be biased at that point. As next study could thoroughly investigate long-term impacts of technology adoption on productivity in various disciplines of economy. A company can foster productivity of its workforce and boost performance by making available to them friendly AI tools as well as by providing them with training. This investigation contributes to understanding how AI technology can increase organizational performance in a significant way according to theoretical frameworks such as TAM model and Resource-Based View (RBV).
Artificial Intelligence; Employee productivity; Business Performance; Perceived Usefulness; Customer Satisfaction
Mohamed Yacine DJEMA, Hacene FOUCHAL, Olivier FLAUZAC, LAB-I*, University of Reims Champagne-Ardenne, France
Large language models (LLMs) remain vulnerable to adversarial prompting, yet state-of-the-art certified defenses such as Erase-and-Check (EC) are too slow for real-time use because they must re-evaluate hundreds of prompt variants. We investigate whether a single, attribution-guided deletion can approximate EC’s robustness at a fraction of the cost. Two variants are proposed. Method A keeps an external safety filter but replaces EC’s exhaustive search with one SHAP/feature-ablation pass, erasing the k most influential tokens before a single re-check. Method B removes the filter entirely: we compute SHAP scores inside the generator (Vicuna-7B), excise the top-r% tokens once, and re-generate. On the AdvBench suite with Greedy-Coordinate-Gradient suffixes (|α| ≤ 20), Method A detects up to 75% of attacks when 55% of tokens are removed—two forward passes instead of EC’s linear-to-combinatorial explosion—while SHAP consistently outperforms feature ablation. Method B, guided solely by SHAP, cuts harmful completions from 100% to 5% after deleting the top-20% tokens and sustains single-digit harm rates for 15–45% deletion budgets, narrowing EC’s safety gap yet adding negligible latency. An explainer comparison shows SHAP recovers nearly every adversarial token within the top-5% importance ranks, whereas LIME is slightly noisier and feature ablation trails far behind. These findings expose a tunable speed–safety trade-off: attribution-guided, single-pass excision delivers large latency gains with a bounded drop in worst-case guarantees. Careful explainer choice and deletion budgeting are critical, but attribution can transform explainability from a diagnostic tool into the backbone of practical, low-latency LLM defenses.
Large Language Models, LLMs, Adversarial Prompting, Jailbreak Attacks, Explainable AI, Greedy Coordinate Gradient, Safety Certification and Robustness.
Serena Pei, School of Engineering, MIT, Cambridge, MA, USA
We present a lightweight pipeline using Stable Diffusion v1.5 [7] for generating anatomically accurate brain MRI images depicting tumors. Using a public dataset of 1,426 glioma MRI slices from 233 patients [5,2] we condition image generation on both descriptive text prompts (text input) and visually transformed grayscale MRI slices (visual input). We explore three visual transforms: Gaussian-blurring, checkerboard-masked, and edge-mapped. Inspired by ControlNet [9], our method supports dual conditioning during both training and inference but avoids duplicating the U-Net architecture—significantly reducing memory overhead. This enables training on standard GPUs such as a single 15GB T4 in Google Colab. To assess image realism on synthesized images, we use both qualitative inspection and Fréchet Inception Distance (FID). This model is an important step towards building more flexible, privacy-preserving methods for creating high-quality medical images in low-data, low-memory settings— with potential applications in rare disease research and AI-driven healthcare.
Stable Diffusion, ControlNet, Healthcare, Medical Imaging.
L. De Grandis1, 2, F. Granata1, 2, A. Lanza2, D. Costa2, E. Oro3, and M. Ruffolo2, 1University of Modena, 2Altilia.ai, 3National Research Council of Italy
The development of Large Language Models (LLMs) has opened impressive possibilities in text summarization, enabling the automatic generation of increasingly human-like synopses. However, modern documents are increasingly complex in both structure and content, with the inclusion of numerous semantically complex charts and tables. This complexity is particularly pronounced in financial documents, which pose challenges in multimodal fusion, alignment, and coherence, with users expecting information to be integrated from and presented across multiple modalities. This work proposes a pipeline that integrates document understanding (DU) models and prompting strategies as a solution for long financial document summarization. Moreover, through a markdown-structured document representation and the use of carefully designed prompt templates, we summarize the document’s content and augment it with images and tables, effectively achieving multimodal summarization with multimodal outputs (MSMO). Experimental results highlight improved factual accuracy and effective inclusion of multimodal information, advancing MSMO.
Long Document Summarization, Multimodal Document Summarization, Large Language Models, Financial Document Analysis.
F. M. Granata1, 2, L. De Grandis1, 2, A. Lanza2, D. Costa2, E. Oro2, and M. Ruffolo2, 1University of Modena, 2Altilia.ai, 3National Research Council of Italy
We propose an approach for leveraging large language mo-dels (LLMs) to answer questions over tables. Our work investigates both direct question answering and semantic parsing paradigms by converting natural language queries into SQL queries, Pandas code, and vanilla Python functions. Evaluations conducted on multiple benchmark datasets including FinTabNetQA, VWTQ, VTabFact, and a proprietary financial dataset (FinTab-It) reveal that HTML table representations enhance model performance, with GPT-4o exhibiting consistent accuracy and LLama 3.1 8b demonstrating sensitivity to input format. Furthermore, fine-tuning LLama 3.1 using QLoRA in low-resource settings yields modest performance improvements. These findings highlight the potential of LLMs to simplify and improve table-based question answering, and they open avenues for future research on optimized fine-tuning and alternative intermediate representations. .
Large Language Models, Question Answering, Tables, Semantic Parsing, HTML.
Salman Bader Hazza, Ibrahim ALhajouj, Abdullah Nader Aldossary, Moteb Abdullah Aldossary, and Saud Alhajaj Aldossari, Department of Electrical Engineering, Prince Sattam bin Abdulaziz University Wadi Aldawaser-11913, Saudi Arabia
With the upcoming farming revolution, this paper presents the development of an AI-powered agricultural monitoring system that integrates IoT devices with machine learning algorithms for real-time soil data analysis and nutrient prediction. A custom-built sensor-based device was designed to collect environmental data, including temperature, humidity, and essential soil nutrients (Nitrogen, Phosphorus, and Potassium). The collected data was preprocessed and used to train various supervised learning models, including Neural Networks, Random Forests, and CatBoost. These models were evaluated using key regression metrics such as MSE, MAE, and R2 to determine their predictive accuracy. The results demonstrate that AI techniques can significantly enhance nutrient estimation and decision support in precision agriculture. This study contributes to the growing field of smart farming by offering a low-cost, sensor-integrated solution for sustainable agricultural monitoring.
IoT, AI, catboost, random forest, neural network, farming technologies
Maher Rebai1 and Taha Houda2, 1De Vinci Higher Education Av. L´eonard de Vinci, 92400 Courbevoie, France, 2Price Mohammad Bin Fahd University 617, Al Jawharah, Khobar, Dhahran 34754, Arabie saoudite
Efficient path planning for mobile sensors is crucial in Wireless Sensor Networks (WSNs) to ensure optimal monitoring of coverage holes while considering real-world constraints. This work addresses the problem of determining an optimal trajectory for mobile sensor ensuring optimal monitoring of coverage holes while efficiently navigating through the sensing field. We introduce a novel Binary Integer Linear Programming (BILP) model that formulates the trajectory planning problem as a discrete optimization task, allowing for fine-grained control over sensor movement and coverage quality. The performance of the proposed approach is thoroughly evaluated through comparative experiments against both exact and heuristic methods from the literature. The obtained results confirm that the proposed approach outperforms the recent existing methods.
Wireless Sensor Network (WSN), Linear programming.
Ahmed Khan Leghari1, Mads Johansen1, and Andreas Lyndrup Jensen2, 1Digital & Sustainable Innovation, FORCE Technology, Denmark, 2Liquid Flow & Type Approval, FORCE Technology, Denmark
The exponential growth of IoT applications and related hardware has dramatically changed the manual way of doing things. Metrology is one such field that is witnessing widespread changes caused by digitalization and connectivity of metrological hardware. Metrological devices need calibration at regular intervals to maintain their metrological accuracy. After the calibration, each device gets a calibration certificate that documents the findings of the calibration and is valid until the next calibration. These calibration certificate are issued on a physical paper or as a PDF documents to be human readable. However, efforts are under way to come up with a globally acceptable machine readable calibration certificate called DCC (Digital Calibration Certificate). Growing demand of digitalization and connectivity of metrological devices means that the paper and PDF based calibration certificates will be soon thing of the past, and will eventually be replaced by machine readable Digital Calibration Certificates (DCCs). DCCs when transferred over the network could be exposed to security threats such as man-in-the-middle (MITM) attack. This paper propose a DCC exchange model that ensures integrity, confidentiality, non-repudiation and authenticity of DCCs sent over network from a calibration lab to the organization that requested for the equipment calibration.
Digital Calibration Certificate, DCC, man-in-the-middle (MITM), Calibration, IoT, Security, Integrity, Confidentiality.