Dec' 23
The IUP Journal of Information Technology
Focus
The new data created by GenAI can be of the same data type as the input data, for example, text-to-text; or it can be of a different data type, for example, text-to-image. GenAI has been widely used in various applications, including text generation, image generation, music composition, video creation, dialogue generation, etc. Popular examples of GenAI include ChatGPT, MuseNet, BigGAN and StyleGAN.
There are multiple models of GenAI like generative adversarial networks (GANs), variational autoencoders (VAEs), transformer models, recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), etc. GANs are the most common models used today. They consist of two neural networks-a generator and a discriminator-trained simultaneously through adversarial training. The generator creates new data, while the discriminator evaluates the authenticity of the created data, i.e., decides if it is from the existing content or generated. The competition between the two networks results in the generator improving its ability to produce realistic content. VAEs are a type of generative model that focuses on learning the underlying distribution of the input data. They consist of two neural networks, typically referred to as encoder and decoder. The encoder maps input data to a latent space and the decoder reconstructs the input data from the latent space. They are particularly used for generating diverse and structured content. RNNs and LSTMs are sequential models that can be used for generating sequences of data, such as text or music. They can capture dependencies and relationships within sequential data, making them suitable for tasks like language modeling and text generation. Transformer models, such as generative pre-trained transformer (GPT) have gained prominence for their ability to handle sequential data efficiently. GPT, for example, is a large-scale language model that can generate coherent and contextually relevant text based on the input it receives. GenAI models are still in their early stages and need to grow in terms of computing power needed, speed, ethics and unbiasedness.
The first paper, "Revolutionizing Image Captioning: A Fresh Perspective Through Stylistic Enhancement and Adversarial Learning", by Sushma Jaiswal, Harikumar Pallthadka, Rajesh P Chinchewadi and Tarun Jaiswal, presents an approach to generate automated captioning of images using the Attention-GAN model. With qualitative and quantitative analysis, the authors have stated that the approach is a promising image captioning approach.
The second paper, "Green Cloud Model for Improved Healthcare Information and Collaboration in Ethiopia", by Temesgen Tona, Getachew Tewachew and Durga Prasad Sharma, suggests a cloud-based green information process modeling for federated healthcare institutions.
The last paper, "Risk Mitigation Using Robotics: Case Studies", by Venkata Ravi Ram Pinninti and Pavitra Pinninti, shows that robotic crawlers can be used to eliminate safety-related risks, while significantly reducing the cost of inspections.
Consulting Editor
Revolutionizing Image Captioning: A Fresh Perspective Through Stylistic Enhancement and Adversarial Learning
Attention-GAN is an innovative model for captioning images that combines generative adversarial networks (GANs) with attention mechanisms in a smooth and seamless manner. The proposed model comprises two main parts. In order to prioritize important visual components for contextually rich captions, an attention-based caption generator first creates strong associations between visual areas and caption segments. Second, the introduction of aesthetic variation through an adversarial training process results in refined and styled descriptions that incorporate creative variances as well as content. This dual-component approach generates engaging and diverse image captions by fusing creativity through adversarial learning with accuracy through attention-based modeling. The capacity of Attention-GAN to produce visually beautiful and contextually relevant captions is demonstrated through extensive trials on benchmark datasets. Both quantitative and qualitative analyses validate the model's ability to generate captions that are consistent with image content and accommodate a range of artistic subtleties. For a broad range of computer vision and natural language processing applications, Attention-GAN is a promising technology that bridges the gap between factual description and creative expression.
Green Cloud-Based Model for Improved Healthcare Information and Collaboration in Ethiopia
The paper analyzes the issues and challenges of the current information and communication process of private sector healthcare institutions in Ethiopia. Both qualitative and quantitative research approaches were used to collect and analyze data using surveys, interviews, observation and related secondary source documents. The analysis indicated the research gaps in the current information and communication process models and the clear direction toward a solution. This study employed a mixed research design (both exploratory and applied research) and used Google Forms for a survey and interview-based fact-finding and analysis, Microsoft Visio for model designing, and Protopie for prototyping. As a final contribution, the study designed and validated a cloud-based green information and communication process model for the federated healthcare institutions in developing countries in general and the Southern Nations, Nationalities and Peoples Region (SNNPR) of Ethiopia in particular. This model can be used as an important instrument for inter or intra institutional collaboration and communications. Based on the user acceptance and demonstrated validation, the results indicate improvements in energy, efficiency, unified data repository, intra institutional collaboration, information sharing, trusted and ethical information and unified integration and share-ability. The test validated the model and its prototype as a new knowledge contribution to the domain. The research findings can be used as a basis for next-generation transformation of private sector healthcare institutions in developing countries such as Ethiopia.
Risk Mitigation Using Robotics: Case Studies
Use of drones in plant inspection, in particular inspection involving tall stacks, to reduce safety risks has picked up considerably in the recent past. However, there are essential plant inspections that involve detailed contact-based inspection, which cannot be achieved using drones. Such cases include close-up inspections like ultrasonic inspection for corrosion studies and inspections that require customized tools. In these scenarios, use of robotic crawlers is the superior option. Leaks or accidental release of chemicals in process plants can have severe consequences, not only for workers and neighboring populace but also for the reputation and financial stability of organizations. Thus safety risk management becomes a crucial component in any work undertaken in every industry. Ultrasonic inspections involve nondestructive testing of wall thickness of tanks and piping to assess the risk of failure. Such inspections require elaborate access arrangements to preclude the possibility of fall from height accidents. The optimal way to prevent accidents to eliminate the hazard as far as reasonably practical. When work-related specific hazards are eliminated, the probability of accidents in that work becomes zero. The paper presents two practical case studies implemented in two different industries, where safety-related risk was completely eliminated using robotics, with collateral benefits in the form of reduction in outage times and a significant 60-90% overall costs savings.
Upload Articles
Click here
to upload your Articles
- HRM Review
- Marketing Mastermind
- Global CEO
- The IUPs World of IOT