Abstract:

The innovation of transformer architecture has propelled the growth of Natural Language algorithms and models, spanning language models, large language models, and pre-training image–text foundation models. Integrating these advancements with state-of-the-art Natural Language Processing (NLP) techniques has led to innovative applications with active use. Achieving Artificial General Intelligence in the hyperspectral imaging (HSI) domain necessitates extracting abstract features from images, emphasizing global features over pixel-level local features using spectral and spatial characteristics. Abstract features, extracted by leveraging the abstract nature of natural language, provide more relevant information. Language, with its consideration of sequence modeling, semantic information and image context, proves instrumental in this process. The fusion of NLP's language processing capabilities with HSI's detailed spectral information opens avenues for holistic and context-aware applications. This review explores the convergence of NLP and HSI, highlighting their potential synergy in HSI tasks. The review begins by elucidating the fundamental principles of integrating NLP and HSI, establishing a foundational understanding for readers unfamiliar with the integration. Subsequently, the synthesis of NLP and HSI is examined, with a focus on recent developments, challenges, and potential applications. The review addresses the integration of Sequence Models, Language Models, Transformer Based Architectures, and Image-Text models with Hyperspectral imaging for various tasks including HSI Classification, Feature Extraction, Semantic Segmentation, Image Captioning, Visual Question Answering, Object Detection, Hyperspectral Unmixing, Text–Image Retrieval, Change Detection, Image Reconstruction and Hyperspectral Image Super-Resolution. It critically evaluates the advantages and future research perspective of using Natural Language based architectures and concepts HSI applications and tasks.

Abstract:

The COVID-19 outbreak has disrupted the world, emphasizing the urgent need for efficient drug development. Traditional drug discovery methods, faced with a vast chemical search space of approximately 1060 drug-like compounds, demand extensive time and financial investments. Conventional in-lab techniques allow testing only 105 compounds per day, significantly escalating the cost and duration of drug discovery. To address these challenges, the automation of drug discovery through computational tools and algorithms is crucial. Drug producers are increasingly integrating AI techniques such as Graph Neural Networks, Deep Learning, and Machine Learning to streamline the process. Viewing the molecule structure as a sequence of strings, similar to predicting words in a natural language sentence, has proven effective. Generative Artificial Intelligence (AI) algorithms, particularly those based on language modeling, are adept at efficiently generating data by learning patterns from training data. This paper reviews the algorithms and methodologies of Generative AI in the context of drug discovery. Starting with fundamental concepts and various types of Generative AI models, the review delves into their application in the drug discovery process. The paper scrutinizes existing approaches that utilize Natural Language Based, Generative Adversarial Networks, and Variational Autoencoders Generative AI models, highlighting their features. The review concludes by outlining future research directions, providing insights for leveraging Generative AI models in the field of pharmaceutical sector.