Authors - Aakanksha Jain, Tejaskumar Bhatt, Darshita kalyani, Jatin Modh, Abhishek Jain Abstract - The effectiveness of the deep learning models ResNet50, MobileNetV2, VGG16, InceptionV3, and EfficientNetB0 in identifying the freshness of food is evaluated in this study using visual analysis. In resource-constrained situations, these designs are ideal for automated food quality inspection and real-time freshness monitoring as they offer higher accuracy or computational efficiency. Transform learning are used to develop binary classifiers, which were then trained on a dataset of annotated food photos and assessed for efficiency and accuracy. After undergoing standardized preprocessing, the models' capacity to differentiate between fresh and stale food in a variety of test photos was evaluated. The results show the advantages and disadvantages of each model. The current stream of research frequently concentrates on generic picture classification tasks instead of the particular difficulties of distinguishing subtle visual differences in food freshness, leaving a gap in knowledge of model robustness under real-world conditions. The study advances deep learning applications in food quality evaluation by offering useful insights for model selection based on operational requirements.
Tuesday August 25, 2026 9:30am - 11:30am IST Virtual Room AGOA, India
Authors - Jyotiprakash Mishra, Sanjay K. Sahay, Aman Pathak Abstract - RISC-V is an open-source Instruction Set Architecture (ISA) designed with a modular and extensible structure, allowing for customizable implementations. Its simplified base ISA, combined with optional standard and custom extensions, provides flexibility for a wide range of computing applications, from embedded systems to high-performance computing. Its open design accelerates innovation and customization but also introduces security challenges by exposing the architecture to potential attacks. While RISC-V offers significant advantages, its lack of standardized security features compared to proprietary ISAs like ARM and x86 highlights persistent risks, particularly in security-critical applications. For this reason, scrutiny of RISC-V’s security is crucial due to its widespread use in academia and its adoption by countries like India and China, who are looking to benefit from its open nature. We review the current security challenges in RISC-V, examining key vulnerabilities that exist in areas such as the microarchitecture, trusted execution environments, secure enclaves, secure boot, cryptographic instruction set architectures, memory encryption, and electromagnetic fault injection attacks. This review aims to cater to the needs of modern researchers for the development and implementation of the RISC-V ISA in a secure manner.
Authors - Nithin Kandi, Murari Nallamalli, Dorai Sai Charan M, Vijay G, Beena B. M. Abstract - Dynamic load balancing in distributed computing environments, especially with heterogeneous nodes, remains a significant challenge due to the fluctuating nature of workloads and resource availability. This paper presents a novel approach leveraging Deep Deterministic Policy Gradient (DDPG), a reinforcement learning algorithm, for optimal workload allocation in real-time systems. The system aims to minimize latency and maximize resource utilization by dynamically adapting to varying node metrics, including CPU usage, memory load, and latency. The DDPG model is trained on simulated state data, and real-time inference is performed through an API Gateway, enabling seamless integration with a five-node cluster. Results demonstrate that the proposed system outperforms traditional static and heuristic approaches in balancing workloads, optimizing resource utilization, and reducing latency. The approach is scalable, robust, and easily adaptable for edge and hybrid cloud architectures, providing a cost-effective solution for dynamic load balancing in distributed systems. This work bridges the gap between traditional cloud infrastructure and edge computing, ensuring efficient resource management in real-time systems.
Authors - Jyotiprakash Mishra, Sanjay K. Sahay, Aman Pathak Abstract - Modern processors have achieved significant performance enhancements through the implementation of speculative execution. These enhancements stem from hardware optimizations that not only improve performance but also introduce side channels, which are exploited to undermine the system’s security model. Following the discovery of Spectre, which revealed that speculative execution pipelines could bypass security boundaries, nearly all microarchitectural structures and hardware optimizations have become targets for exploitation. In the wake of these attacks, the immediate response from organizations releasing mitigation patches led to noticeable performance degradation overnight, without fully addressing the underlying issues. This paper provides a comprehensive review of architecture-agnostic attacks on modern computing systems, tracing their evolution from the initial emergence of Spectre to contemporary attacks targeting Apple Silicon. We detail the mechanisms of these attacks, the environments in which they are exploited, and their broader security implications. Furthermore, we analyze various mitigation strategies that have been proposed, acknowledging that these strategies often fail to fully resolve the issues and typically incur a performance cost. These mitigation patterns include proposed changes in operating systems, hardware, and compilers. This review aims to provide researchers and architects with the foundational knowledge needed to develop more effective mitigation strategies that address the vulnerabilities while minimizing performance overhead.
Authors - Ankit Agarwal, Ankur Pandey, Ashlesh Kumar, Dhanush D, Swetha G Abstract - Steady-State Visual Evoked Potential (SSVEP)-based Brain Computer Interfaces (BCIs) are a promising tool for non-invasive neural communication and control, particularly for individuals with severe physical or medical conditions that limit conventional interaction. However, accurately detecting and classifying SSVEP signals remains challenging due to noise and inter-subject variability. This study evaluates the performance of established classification methods, including Canonical Correlation Analysis (CCA), Filter Bank CCA (FBCCA), and transfer learning models such as EEGNet, DeepConvNet, and ShallowConvNet. To address the limitations of existing methods, we propose a novel hybrid approach combining CCA, Continuous Wavelet Transform (CWT), and Convolutional Neural Networks (CNN). This method aims to enhance feature extraction and classification accuracy. The models were evaluated on the benchmark SSVEP dataset from Tsinghua University, with preprocessing steps involving independent component analysis (ICA) and band-pass filtering. FBCCA achieved the highest accuracy of 97.5%, followed by CCA (93%) and DeepConvNet (86.95%). Our proposed method attained an accuracy of 77.52%, demonstrating its potential for robust SSVEP classification. These results underline the value of advanced algorithms and preprocessing strategies in improving SSVEP-based BCI performance, paving the way for more effective assistive technologies.
Tuesday August 25, 2026 9:30am - 11:30am IST Virtual Room AGOA, India
Authors - Radhika V. Kulkarni, Avish Agrawal, Aryan Vimal, Rohan Barde, Raghav Bajaj, Khursheed Gaddi Abstract - The Indian judicial system heavily relies on precedents for legal interpretations and decision making, providing access to relevant case law a critical yet time consuming task for legal professionals and researchers. This paper presents an AI-powered Legal Case Search Engine designed to transform legal re-search by leveraging advancements in Natural Language Processing (NLP) and Large Language Models (LLMs). The system enables efficient retrieval of contextually relevant legal precedents from the Supreme Court of India’s judgments, utilizing techniques like vector embeddings, cosine similarity, and semantic search. It offers concise case summaries and metadata insights to streamline decision making and improve accessibility to legal data. Built on open-source technology, the platform emphasizes scalability, cost efficiency, and user centric design, ensuring adaptability for future enhancements like multilingual support. By democratizing access to legal knowledge, this research aims to bridge the gap between complex legal texts and their practical application, fostering innovation in legal workflows and enhancing the rule of law.
Authors - Prajwal S, Praveen M P, Dhanya M Abstract - The global transition toward electric mobility is crucial in mitigating climate change, reducing air pollution, and promoting sustainable urban transportation. India, one of the fastest-growing markets for electric vehicles (EVs), has witnessed a surge in electric two-wheeler (E2W) adoption. However, concerns regarding battery longevity, charging infrastructure, and affordability remain key barriers to widespread adoption. This study applies sentiment analysis techniques to assess consumer perceptions of electric bikes using machine learning models for sentiment classification. A dataset comprising 3,395 consumer reviews was collected from leading automotive platforms, including BikeWale, BikeDekho, OneDrive, and ZigWheels, using web scraping techniques. The data was analyzed using VADER, TextBlob, Naïve Bayes, Logistic Regression, and Support Vector Machines (SVM) to classify sentiment and identify key consumer concerns. The results indicate a predominantly positive sentiment towards electric bikes, driven by environmental benefits and cost savings. However, consumers expressed concerns over battery efficiency, charging station availability, and high initial costs. Among the models tested, SVM achieved the highest accuracy, making it the most effective in sentiment classification. This study contributes to the limited academic research on electric bikes, offering data-driven insights into consumer perceptions. By utilizing real-world consumer data from widely used automotive platforms, the research provides valuable information for policymakers, manufacturers, and industry stakeholders. The findings aim to assist in developing strategies to address consumer concerns and enhance electric bike adoption in India.
Authors - Vani E S, Akshay Sinha, Rahul Singh Rajput, Pranjal Krishna Gupta, Chiranth K M Abstract - The primary goal of any educational institution is to provide students with a high-quality learning experience and comprehensive knowledge. Identifying students who need additional support and implementing effective strategies to enhance their academic performance is critical to achieving this objective. This study applies three machine learning techniques to develop a predictive model for assessing student performance across various academic disciplines and institutions. The techniques include Logistic Regression, k-Nearest Neighbours (KNN), and Support Vector Machine (SVM). The models were evaluated using metrics such as the Receiver Operating Characteristic (ROC) index and classification accuracy. Additional performance indicators, including classification error, precision, recall, and the F1-score were also computed. The dataset, which consists of data from a student survey and academic records, included information from a total of 700+ students. Among the models tested, the SVM model outperformed the others, achieving an ROC index of 0.82 and a classification accuracy of 84.04%.
Tuesday August 25, 2026 9:30am - 11:30am IST Virtual Room AGOA, India
Authors - Bhadouriya Khushi Mukeshsingh, Rajput Adityasingh Shashikantsingh, Parmar Smit Dharmeshkumar, Tiwari Prashant Dineshkumar, Nirav D. Mehta, Anwarul.M.Haque Abstract - Fire emergencies pose significant risks, with conventional alarms often lacking rapid response mechanisms. Delays in manual intervention can lead to severe consequences in residential and industrial settings. This study presents the Domestic Emergency System (DES), an IoT-integrated, multi-layered fire detection and emergency response framework. DES utilizes flame and smoke sensors, GSM-based emergency dialing, and RF communication modules to enable real-time hazard detection, structured evacuation, and automated alerts. An adaptive thresholding algorithm minimizes false alarms while ensuring precise fire identification. Experimental validation demonstrated fire detection within 0.8 seconds, RF signal propagation up to 25 meters, and emergency call execution within 5 seconds, improving response time by 70% compared to traditional alarms. The multi-tier alert mechanism and mobile notifications enhances situational awareness and security coordination. The DES framework outperforms conventional fire alarm systems by integrating automated emergency calling, RF-enabled intra-building communication, and a resilient multi-modal alert system. Its scalability and adaptability to smart city infrastructure make it a transformative solution for modern safety applications.
Authors - Samyuktha D, Chandrathara P T, Gowri Gireesh, Rojalin Patri Abstract - Financial literacy affects individuals' financial conduct, especially their capacity to save, consume well, and engage in thrift behaviors. In this study, the connection among financial literacy and various thrift behaviors—Careful Thrift, Spending Thrift, Saving Thrift, and Lifestyle Thrift—amongst college students is examined. Utilizing a survey quantitative methodology, statistical analysis is conducted to assess how financial literacy can predict such thrift behaviors. It also explores if demographic variables of gender and financial support sources are moderators of such relationships. Results show that financial literacy positively influences cautious thrift, lessens irresponsible spending, and increases saving and lifestyle thrift. The paper finally ends with suggested recommendations for financial literacy programs used to enhance financial decision-making skill among students.
