http://localhost:8081/jspui/handle/123456789/15626 2026-04-23T06:03:32Z http://localhost:8081/jspui/handle/123456789/18874 Title: INTEGRATED REAL-TIME TRAFFIC SIGN MONITORING AND ANOMALY DETECTION Authors: Pal, Tirtharaj Abstract: Road signs are critical elements of road safety, but anomalies in these signs can lead to confusion and accidents. This work presents a multi-stage approach to detecting and rectifying anomalous road signs in the Indian context. Leveraging deep learning algorithms like YOLOv8 and innovative methodologies such as synthetic data generation and autoencoder-based anomaly detection, we aim to enhance road sign management systems' accuracy and efficacy. We collected data along various routes, trained YOLOv8 with both normal and synthetic anomalous signs, and employed SAM for segmentation. Our results demonstrate promising performance metrics, indicating the effectiveness of our approach. Future work involves expanding the scope to include more anomalies and intensive autoencoder training, ultimately contributing to global road safety and the future of autonomous vehicles. 2024-05-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/18873 Title: MINIMIZING ENERGY USAGE FOR CAPACITATED VEHICLE ROUTING PROBLEM USING POLICY GRADIENT (RL) AND COMPARING SOLUTIONS WITH GUROBI Authors: Kumar, Shivam Abstract: The Vehicle Routing Problem (VRP) is a well-known optimization challenge with practical applications in various areas, including delivery route optimization, transportation logistics, and mobile resource allocation. This work specifically focuses on the capacitated vehicle routing problem (CVRP), where each customer has a demand that must be met by vehicles with capacity constraints. The goal is to minimize the total cost or distance traveled by vehicles while ensuring that each customer’s demand is satisfied and vehicles stay within their capacity limits. Traditional methods for solving CVRP typically use heuristics and mathematical programming techniques. However, these approaches can struggle with large-scale instances and dynamic environments. In recent years, reinforcement learning (RL) has emerged as a promising solution for complex optimization problems. In this study, we use RL techniques to address the CVRP. Our approach harnesses the power of deep RL algorithms, specifically combining deep neural networks and actor-critic methods, to learn effective policies for route planning and optimization. By framing the CVRP as a Markov Decision Process, we develop an RL agent that learns to make sequential decisions regarding vehicle movements and load allocations. We evaluate our proposed RL framework on a set of benchmark VRP instances, comparing its performance against the state-of-the-art solver Gurobi. The experimental results demonstrate that our approach achieves competitive solution quality and computational efficiency, especially in larger problems. Additionally, we explore the robustness and generalization capability of the learned policies by assessing their performance on unseen problem instances and assessing the performance of RL(trained for larger problems) on smaller problem instances. Overall, our work underscores the potential of reinforcement learning as a promising methodology for solving the Vehicle Routing Problem. Through the integration of deep RL algorithms and problem-specific insights, we illustrate that RL can provide efficient and effective solutions to this challenging optimization problem, opening avenues for further advancements in the field of transportation logistics and route planning. 2024-06-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/18872 Title: ENHANCING FAKE-NEWS DETECTION WITH ADAPTIVE RATIONAL GUIDANCE USING LLM(GPT-4O, GPT-3.5, CLAUDE-3.5) GENERATED FEATURES Authors: Bari, Sayyad Abdul Abstract: Fake-news poses significant threats to society, affecting public opinion, political stability, and public health. Effective detection of fake-news is crucial, yet challenging, due to its complex and evolving nature. This thesis explores the enhancement of fake-news detection using an Adaptive Rational Guidance Network (ARGN), which integrates rationales generated by advanced Large Language Models (LLMs) such as Claude 3.5 Sonnet , GPT-4o and GPT-3.5. Initially, we develop a baseline of ARGN model, utilizing content-only data for prediction. Subsequently, rationales were generated for the same data through textual and commonsense analyses conducted by different LLMs. These rationales were then incorporated into the ARGN model, allowing for a comparative analysis of models using content-only data versus those using both content and LLM-generated rationales. The results demonstrated a notable enhancement in fake-news detection accuracy when rationales were included. The thesis presents a comprehensive evaluation of the ARGN model's performance across various metrics, including F1-score, Accuracy, SPAUC, Precision & Recall. Results indicate that incorporating LLM-generated rationales enhances model performance, giving deeper insight into and ability to recognize false information. Claude-3.5 emerged as the most effective model, consistently outperforming GPT-4o and GPT-3.5 across multiple metrics. This research underscores the potential of integrating advanced LLMs with adaptive models to improve fake-news detection. The proposed ARGN model not only enhances detection accuracy but also offers a robust framework for leveraging multi-perspective analyses in natural language processing tasks. Future work could explore further optimization and application of this model in diverse domains requiring high-stakes decision-making. Keywords: Fake news detection, Adaptive Rational Guidance Network, Large Language Models, BERT, Co-Attention Networks, Accuracy, Precision, Recall, AUC, Machine Learning, Online Misinformation, Text Classification, Natural Language Processing, ARG Model, Adaptive Attention Mechanisms, GPT-4o, GPT-3.5, Claude-3.5, rationales, textual analysis, commonsense analysis, model performance, F1 score. 2024-05-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/18870 Title: WORD EMBEDDING IMPROVEMENTS IN NATURAL LANGUAGE PROCESSING Authors: Kumar, Rahul Abstract: The growing intricacy of problems related to natural language processing has brought to light the shortcomings of static word embeddings in terms of encapsulating sentence context. Transformer-based models that use self-attention processes have been used to solve this, however they have problems because of BERT's tokenizers word piece algorithm, which divide words into sub-words and may lead to problems with contextual meaning understanding. By improving the tokenizer, this research aims to improve vector embeddings of words. The tokenizer was altered with domain-specific terms or tokens, and the model was trained on a dataset that contained these tokens. According to our research, the prepared words' cosine similarity has improved, suggesting improved contextual representation. These stretching of word embeddings imply that certain words have better embedding quality than previously existing models, with some words having more accurate vector space representations. Furthermore, our methodology emphasizes the significance of tailoring tokenization techniques to particular domains, which may result in more accurate language models dynamic embeddings. This research shows the problem of existing tokenizer which are word piece based algorithm. By adding the special tokens into the tokenizers vocabulary and it shows the significant difference of dynamic embeddings generated by the BERT model. 2024-06-01T00:00:00Z