Aamir Wali

Rice is the main staple food of Pakistan. The rising demand for rice, driven by population growth, is under threat by the bacterial diseases such as leaf blight, leaf blast, and brown spot. These diseases adversely affect rice production, reducing yield and threatening food security. Technological advancements, particularly in deep learning, offer opportunities to strengthen agricultural practices in countries like Pakistan. Early and accurate detection of these diseases is critical to ensuring food security and preventing potential shortages. In this research, we propose a novel pipeline for rice disease detection using a hybrid approach that integrates Generative Adversarial Networks (GANs) for data augmentation, Otsu thresholding for image preprocessing, and a deep learning model for classification. While existing models like Inception V3 and MobileNet V2 have shown promising results on training datasets, their performance diminishes on larger, more diverse datasets, due to overfitting and poor generalizations. Our proposed hybrid pipeline aims to address rice disease detection challenges by integrating (i) synthetic data generation using a Pix2Pix GAN to augment the dataset, (ii) a pre-processing stage employing Otsu thresholding on the augmented images to enhance features, and (iii) a final classification stage using an InceptionDenseNet model trained on the processed data. Our suggested approach achieved an accuracy of 86.98%, a significant improvement from the baseline accuracy of 74% obtained by the Inception DenseNet model on the original dataset. This enhancement demonstrates the effectiveness of the combined GAN-based augmentation and Otsu thresholding pre-processing.

This study systematically evaluates the performance of state-of-the-art large language models (LLMs) in medical and healthcare applications, focusing on their accuracy in answering domain-specific questions. Using benchmark medical question-answering datasets-PubMedQA, MedQA, and MedMCQA-we assess a diverse set of LLMs, including GPT-4, Med-PaLM-2, OpenBioLLM, BioMistral, MediTron, MedAlpaca, and AlpaCare. Our analysis highlights the varying capabilities of these models across different datasets, emphasizing the impact of model scale, domain-specific fine-tuning, and dataset-specific challenges. Larger models such as OpenBioLLM-70B and Med-PaLM-2, consistently outperformed smaller models, showcasing the benefits of extensive training data and computational resources. However, smaller models, like BioMistral-7B, demonstrated competitive performance on specific datasets, suggesting their potential for resource-constrained environments. Beyond accuracy, we explore the broader implications of LLMs in healthcare, including their applications in medical diagnosis, patient care, clinical decision support, and drug discovery. Despite their promise, LLMs face critical challenges, such as the need for explainability, robust data security, bias mitigation, and hallucination reduction. We conclude that while challenges remain, LLMs hold significant potential to transform healthcare by enhancing efficiency, improving patient outcomes, and facilitating advancements in medical research. Addressing the limitations and promoting responsible innovation will be essential to unlocking their full potential for a patient-centered and equitable healthcare future.

Automatic text summarization (ATS) has seen considerable development, but state-of-the-art models depend on vast training data, limiting their applicability in scholarly communication and digital knowledge management for technical and domain-specific texts where annotated datasets are scarce. This article addresses this critical gap. First, to create a controlled, data-constrained setting for study, we created two specialized datasets: CS-Summ and Geog-Summ, comprising 794 and 438 manually summarized excerpts from the Computer Science and Geography domains, respectively. We then propose and evaluate a simple, computationally inexpensive method of keyword-guided fine-tuning to enhance performance. By prepending extracted keywords, identified using spaCy’s NER model, to the source document as an explicit signal of salience, we guide the model’s attention during generation. Using these datasets, we conduct baseline experiments by keyword-guided fine-tuning on T5, BART, and BERT models and comparing their performance against zero-shot GPT-4. Evaluation using standard lexical metrics demonstrates that our keyword-guided approach consistently improves performance over the fine-tuned baselines. More critically, using factual consistency metrics such as BERTScore and NLI-based metrics for abstractive summarization, we show that keyword guidance reduces model hallucination and produces more reliable and factually accurate summaries. Our findings advance computational methods for technical content summarization and underscore the practical value of this method in the development of more robust and accessible summarization systems in resource-constrained environments, contributing to broader efforts to develop digital tools that support specialized scholarly analysis.

Generative artificial intelligence (AI) is being used to generate all kinds of digital data: images, speech, text, and so on. with high fidelity. Within natural language processing, generative AI has many interesting applications such as summarization and story generation. In this article, we focus on generating alternate endings for short stories. Although some work has been done on story generation, to the best of our knowledge, very little work has been done on generating alternate endings. For a more precise ending, a large amount of context needs to be provided making story ending generation a more challenging problem than complete story generation. We propose a complete pipeline consisting of text understanding and text generation modules for generating story endings. Within story understanding modules, we propose and integrate various sub-modules to capture the context of the story. This context is combined with the original story during the training and story generation phase for which a T5 transformer is used. Compared to previous studies, our proposed model reports a higher BLEU value. We also conducted a subjective user study to evaluate the endings generated by our model. Machine-generate endings were given an average score of 3.82 (out of 5) as opposed to 3.86 for human-generated endings.