Inception Institute of Artificial Intelligence

Abstract Artificial Intelligence (AI) is rapidly transforming the oil and gas (O&G) industry, with the emergence of Agentic AI marking a major leap forward. Unlike traditional rule-based analytics or static machine learning models, Agentic AI refers to autonomous AI agents capable of perceiving their environment, learning continuously, and taking independent actions to achieve defined objectives — all with minimal human intervention. These agents are not limited to isolated tasks; instead, they can operate across the entire O&G value chain, from upstream exploration to downstream refining and distribution, making decisions and executing operations in real time. A pioneering example of this shift is ADNOC's recently launched ENERGYai platform — the industry's first large-scale deployment of Agentic AI. ENERGYai integrates specialized autonomous agents across ADNOC's operations, performing tasks such as seismic interpretation, reservoir management, production optimization, and refining process monitoring. These agents combine domain-specific intelligence with the reasoning capabilities of large language models (LLMs), such as OpenAI's GPT, to enhance decision-making and workflow integration. By leveraging commoditized foundation models and AI orchestration frameworks, ENERGYai demonstrates how O&G companies can achieve significant operational efficiencies, reduce downtime, and improve safety. Agile AI-driven teams are increasingly able to outperform traditional legacy systems by responding to real-time data and conditions, creating a competitive edge through automation and intelligent system coordination. This white paper explores the transformational potential of Agentic AI across upstream, midstream, and downstream segments. Using ENERGYai as a real-world case study, we delve into its applications in seismic data analysis, production forecasting, and autonomous process control. We also examine how combining LLMs with autonomous agents accelerates AI adoption and enables broader organizational agility. However, the adoption of Agentic AI is not without challenges. Data quality, QA/QC protocols, inconsistent data standards (e.g., the need for alignment with the OSDU Data Platform), and concerns around trust and explainability remain key hurdles. This paper addresses how leading O&G players are navigating these issues to unlock the full value of autonomous agents while ensuring operational transparency, governance, and stakeholder trust. In summary, Agentic AI represents a transformative shift in the digitalization of O&G, offering a scalable, intelligent, and adaptive future for energy operations.

ABSTRACT Metachromatic leukodystrophy (MLD) is an autosomal recessive neurodegenerative disorder caused by mutations in the arylsulfatase A (ARSA) gene, resulting in lower sulfatase activity and the toxic accumulation of sulfatides in the central and peripheral nervous system. Children account for 70% of cases and become progressively disabled with death occurring within 10 years of disease onset. Gene therapy approaches to restore ARSA expression via adeno-associated viral vectors (AAV) have been promising but hampered by limited brain biodistribution. We report the development of a novel capsid AAV.GMU01, demonstrating superior biodistribution and transgene expression in the central nervous system of non-human primates (NHPs). Next, we show that AAV.GMU01-ARSA treated MLD mice exhibit persistent, normal levels of sulfatase activity and a concomitant reduction in toxic sulfatides. Treated mice also show a reduction in MLD-associated pathology and auditory dysfunction. Lastly, we demonstrate that treatment with AAV.GMU01-ARSA in NHPs is well-tolerated and results in potentially therapeutic ARSA expression in the brain. In summary, we propose AAV.GMU01-ARSA mediated gene replacement as a clinically viable approach to achieve broad and therapeutic levels of ARSA.

Medical image captioning plays an important role in modern healthcare, improving clinical report generation and aiding radiologists in detecting abnormalities and reducing misdiagnosis. The complex visual and textual data biases make this task more challenging. Recent advancements in transformer-based models have significantly improved the generation of radiology reports from medical images. However, these models require substantial computational resources for training and have been observed to produce unnatural language outputs when trained solely on raw image-text pairs. Our aim is to generate more detailed reports specific to images and to explain the reasoning behind the generated text through image-text alignment. Given the high computational demands of end-to-end model training, we introduce a two-step training methodology with an Intelligent Visual Encoder for Bridging Modalities in Report Generation (InVERGe) model. This model incorporates a lightweight transformer known as the Cross-Modal Query Fusion Layer (CMQFL), which utilizes the output from a frozen encoder to identify the most relevant text-grounded image embedding. This layer bridges the gap between the encoder and decoder, significantly reducing the workload on the decoder and enhancing the alignment between vision and language. Our experimental results, conducted using the MIMIC-CXR, Indiana University chest X-ray images, and CDD-CESM breast images datasets, demonstrate the effectiveness of our approach. Code: https://github.com/labsroy007/InVERGe

Explaining Deep Learning models is becoming increasingly important in the face of daily emerging multimodal models, particularly in safety-critical domains like medical imaging. However, the lack of detailed investigations into the performance of explainability methods on these models is widening the gap between their development and safe deployment. In this work, we analyze the performance of various explainable AI methods on a vision-language model, MedCLIP, to demystify its inner workings. We also provide a simple methodology to overcome the shortcomings of these methods. Our work offers a different new perspective on the explainability of a recent well-known VLM in the medical domain and our assessment method is generalizable to other current and possible future VLMs.

Zero-shot learning (ZSL) aims to classify unseen samples based on the relationship between the learned visual features and semantic features. Traditional ZSL methods typically capture the underlying multimodal data structures by learning an embedding function between the visual space and the semantic space with the Euclidean metric. However, these models suffer from the hubness problem and domain bias problem, which leads to unsatisfactory performance, especially in the generalized ZSL (GZSL) task. To tackle such a problem, we formulate a discriminative cross-aligned variational autoencoder (DCA-VAE) for ZSL. The proposed model effectively utilizes a modified cross-modal-alignment variational autoencoder (VAE) to transform both visual features and semantic features obtained by the discriminative cosine metric into latent features. The key to our method is that we collect principal discriminative information from visual and semantic features to construct latent features which contain the discriminative multimodal information associated with unseen samples. Finally, the proposed model DCA-VAE is validated on six benchmarks including the large dataset ImageNet, and several experimental results demonstrate the superiority of DCA-VAE over most existing embedding or generative ZSL models on the standard ZSL and the more realistic GZSL tasks.

Abstract Recently, Medical Visual Question Answering (VQA) became an active area of research with the induction of several publicly available benchmark datasets and the organization of challenges. Like many competitions, the quest for success has driven the use of increasingly complex neural networks. Winning strategies generally leverage multi-scale architectures and model ensembling to achieve state-of-the-art performance. However, several studies have established the capability of simpler architectures in learning more meaningful features and avoiding over-parameterization. Specifically, the use of MixUp based image augmentation with a simple VGG16 network helped achieve significant improvement in performance for medical VQA. Inspired by this finding, we propose a modified version, VQAMixUp, that leverages both images and questions for augmenting VQA datasets. VQAMixUp combined with a few enhanced training strategies help simple models (with \(\approx 65\)% reduced parameters) achieve state-of-the-performance on benchmark ImageCLEF-VQA-MED validation datasets.KeywordsMedical VQADeep learningMixUpImageCLEF