Recent News

  • Dr Gavaskar Publishes Patent on Generating Prompts December 26, 2024

    The Department of Computer Science and Engineering is proud to announce that Dr S Gavaskar has published his patent titled A System for Generating Prompts for Generative Artificial Intelligence (AI) Applications (Application Number: 202441091788). This groundbreaking invention will prove to be a significant advancement in the field of AI, enabling the creation of contextually rich and user-specific prompts, thereby enhancing the accuracy and usability of generative AI systems across various domains.

    A Brief Abstract:

    This research invention relates to Prompt creation process for Generative AI Application by introducing Four way corpus directory(FWCD) comprising the Persona corpus, Localized Application Specific Content Corpus, Annotator corpus, and Stopword corpus to create well-formed, contextual prompts for AI models. It also employs Semantic Based Categorization and Ranking(SBCR) for semantically categorizing and ranking the content present in the Localized Application Specific Content Corpus . The invention improves the interaction between users and Generative AI. It helps deliver more accurate, semantic based outputs from the AI models, improving the overall performance and usability of the system.

    Explanation in Layperson’s Terms:

    Prompt engineering is the process of using natural language to create instructions that generative artificial intelligence (AI) models can understand and interpret.In this research we have create a system for prompt creation by which the users can create their own prompt with the combination of their persona details,stopword,annotator content from their localized corpus directory before applying to the LLM models such as ChatGPT,Copilot etc.

     

    Practical Implementation / Social Implications:

    • This concept can be implemented in Education Institutions to generate tailored prompts for learning materials and academic projects.Enterprises can use localized content to generate role-specific prompts. and marketing organizations can use to create prompts aligned with specific campaigns or audience demographics.It can also be used in organizations where multiple persons with different roles and responsibilities are there and they have their own localized content for which a prompt has to be created for an Generative AI Application.
      This invention also lets people from different skill levels access the system and create their own prompt for their applications.

    Future Research Plans:

    Future research focus is on creating LLM and AI related applications to the field related to education.cyber security,Legal and Enterprises.

     

    Continue reading →
  • A Unified Learning Framework for Detecting Cyberattacks in IoT Networks December 23, 2024

    The Department of Computer Science and Engineering is proud to announce that Assistant Professors Dr Kshira Sagar Sahoo and Dr Tapas Kumar Mishra, along with their research scholar Ms Arati Behera, have published their paper “A Combination Learning Framework to Uncover Cyberattacks in IoT Networks” in the prestigious Q1 journal Internet of Things, which has an Impact factor of 6.

    The article addresses IoT security challenges by utilising Software Defined Networking technology and AI. The authors use Genetic Algorithm to select the most important data features and a hybrid deep learning model combining CNN, Bi-GRU, and Bi-LSTM to detect cyber-attacks effectively. Tested on real-world IoT datasets, the system demonstrates superior accuracy, faster detection, and lower resource usage than existing methods, making it a promising solution for securing resource-constrained IoT networks.

    Abstract

    This study addresses the security challenges in IoT networks, focusing on resource constraints and vulnerabilities to cyber-attacks. Utilising Software Defined Networking and its adaptability, the authors propose an efficient security framework using a Genetic Algorithm for feature selection and Mutual Information (MI) for feature ranking. A hybrid Deep Neural Network (DNN) combining CNN, Bi-GRU, and Bi-LSTM is developed to detect attacks. Evaluated on InSDN, UNSW-NB15, and CICIoT 2023 datasets, the model outperforms existing methods in accuracy, detection time, MCC, and resource efficiency, demonstrating its potential as a scalable and effective solution for IoT network security.

    Practical Implementation/ Social Implications of the Research

    The practical implementation of this research lies in enhancing the security of IoT networks, which are increasingly integral to smart homes, healthcare, transportation, and industrial systems. By detecting and mitigating cyber-attacks efficiently, the proposed model can safeguard sensitive data, prevent service disruptions, and ensure the reliability of IoT systems.

    Collaborations

    This research has been conducted in partnership with Umea University Sweden.

    Future Research Plans

    There is potential to enhance the Deep Learning approach further to reduce the execution time at the power crunch device. Additionally, federated learning could be investigated as a use case, especially concerning edge devices within typical software-defined IoT networks.

    Read more

    Continue reading →

TOP