http://210.212.227.212:8080/xmlui/handle/123456789/146 Sun, 17 May 2026 00:00:56 GMT 2026-05-17T00:00:56Z http://localhost:8080/xmlui/bitstream/id/94e30183-02e2-4ffa-bdc9-4170eeef636b/close-up-handwritten-whiteboard-year-168763957.jpg http://210.212.227.212:8080/xmlui/handle/123456789/146 http://210.212.227.212:8080/xmlui/handle/123456789/563 DEVELOPING A CLASSIFICATION MODEL AND SEMANTIC SIMILARITY DETECTION OF SYSTEM AND SOFTWARE REQUIREMENTS USING SENTENCE TRANSFORMERS V R Manasi, Manasan; Adarsh, S Reading and understanding similar type of requirements using human interventions can be a laborious and time-consuming task. However, the complexity of natural language makes it difficult to accurately identify semantic similarities, which makes more difficulty to the task. The majority of algorithms for detecting similarities rely on matching words to words, paragraphs to paragraphs, or the entire page to the matching.. In this project, a novel ap- proach is proposed which uses large language model (LLM) such as Sentence Transformers for detecting semantically similar requirement.The Sentence Transformer models used in this project includes all-MiniLM-L6-v2 and paraphrase-MiniLM-L6-v2. The dataset consists of 10,500 software and system requirements vital for SCANIA’s (German Company) operations. The aim of the project is to find semantically similar system and software requirements and categorize those requirements based on similarity score and compare the performance of Sen- tence Transformers with traditional algorithms such as Word2vec and TF-IDF Vectorizer to find semantically similar requirements. In addition to it a GUI(Graphical User Interface) is built so that user can interact easily and find similar requirements. The project employs several measures to evaluate the performance of the model, including F1 score, precision , accuracy, euclidian distance , cosine similarity.Among ‘all-MiniLM-L6-v2 ‘ and ‘paraphrase- MiniLM-L6-v2’, ‘all-MiniLM-L6-v2’ outperformed better with an accuracy of 92%,precision of 95%,Recall of 82% and F1 score of 88%. Ultimately, the project aspires to revolutionize requirement analysis, driving efficiency and productivity in software and system develop- ment processes through the seamless integration of cutting-edge technologies and intuitive interfaces. Fri, 28 Jun 2024 00:00:00 GMT http://210.212.227.212:8080/xmlui/handle/123456789/563 2024-06-28T00:00:00Z http://210.212.227.212:8080/xmlui/handle/123456789/562 TEXT-DRIVEN HUMAN MOTION SYNTHESIS USING DIFFUSION MODELS K J, Ananthakrishnan; Sumod, Sundar The term ”text-to-motion generation” pertains to the procedure of producing sequences of human motion by using textual input. The work at hand presents considerable challenges, mostly stemming from the wide range of potential motion, the inherent sensitivity of human perception to such motion, and the inherent complexities associated in effectively articulat- ing and characterising it. The existing generative methods for text-to-motion synthesis are characterised by either substandard quality or constrained expressiveness. To overcome this, we introduce, a diffusion model-based framework for text-driven motion generation. The model presents multiple benefits: Firstly, it employs probabilistic mapping to generate mo- tions by refining inputs through denoising processes while introducing variations. Secondly, it features multi-level manipulation capabilities, allowing it to interpret detailed instructions regarding body parts and facilitate the synthesis of motions of various lengths in response to text prompts that change over time. The performance of model is evaluated using metrics such as FID (Fréchet Inception Distance), Diversity, and MultiModality, which measure the quality and diversity of generated motion samples Fri, 28 Jun 2024 00:00:00 GMT http://210.212.227.212:8080/xmlui/handle/123456789/562 2024-06-28T00:00:00Z http://210.212.227.212:8080/xmlui/handle/123456789/561 ESTABLISHING CONSENSUS IN MULTI-AGENT SYSTEMS VIA REINFORCEMENT LEARNING TECHNIQUE Irfan, E; Resmi, R A multi agent system(MAS) in the context of artificial intelligence refers to a system composed of multiple interacting agents, which can be entities or robots. These agents work together or compete to achieve individual or collective goals. Consensus is a decision making process that aims to achieve the agreement of the majority of entities involved, it involves collaborative efforts to reach a decision that is mutually acceptable to all involved. Rein- forcement Learning is an AI algorithm utilized with robots and autonomous vehicles to learn optimal behaviors by interacting with simulated environments, thereby facilitating safer and more efficient real world decision making. In this work, the focus is on the application of RL technique Q-learning using E-puck robots. The E-puck employs Q-learning to deter- mine the most suitable actions to take within its environment. The concept of consensus is examined within the context of multi agent systems, where E-puck robots constitute the agents. Consensus, in this scenario, pertains to the agreement reached among the E-puck agents regarding their relative positions or distances. Establishing consensus among MAS and ensuring efficient communication and coordination among E-puck robots pose significant challenges, yet achieving it holds immense potential for enhancing MAS and optimizing au- tomation and decision making processes. Using RL algorithm along with webots simulation and E-puck robots the frameworks can help different agents to navigate toward a common destination while avoiding obstacles and maintaining consensus. This approach will be ben- eficial for various sectors, including healthcare assistance, defense, warehouse management, and even agricultural automation. Fri, 28 Jun 2024 00:00:00 GMT http://210.212.227.212:8080/xmlui/handle/123456789/561 2024-06-28T00:00:00Z http://210.212.227.212:8080/xmlui/handle/123456789/560 ANALYSING EEG SIGNAL TO RECOGNISE EMOTIONS Dilish, Jose; Mubarak, Ali M Emotion recognition through Electroencephalography (EEG) is widely acknowledged as a reliable method with diverse applications including defense, aerospace, and medical do- mains. This research aims to detect emotions from EEG signals, depicting individuals’ brain activity. Electroencephalogram (EEG) serves as a pivotal tool in Brain-Computer Interface (BCI) systems, capturing brain signals. With the advancement of machine learning algo- rithms and the growing real-world utility of BCIs, the classification of emotions from EEG data has gained prominence. Previously, researchers had limited insights and knowledge into the specific connections between distinct EEG characteristics and various emotional states. This work utilizes the DEAP dataset, featuring 32 EEG recording channels, and employs ensemble models for training. The model’s performance is evaluated using metrics like accu- racy, precision, recall, and F1-score. In this work, we augment traditional feature extraction methods such as mean, standard deviation, entropy, skewness, and kurtosis along with con- tinuous wavelet transformation for enhanced signal analysis, This combined approach aims to capture both basic characteristics and time-frequency domain information within the EEG signals, potentially improving emotion classification accuracy. Fri, 28 Jun 2024 00:00:00 GMT http://210.212.227.212:8080/xmlui/handle/123456789/560 2024-06-28T00:00:00Z