Invited Speakers

Understanding how people perform everyday activities over extended periods can enable new forms of AI-assisted healthcare. For example, intelligent systems in homes could help understand Activities of Daily Living, recognize subtle changes in behavior, and support the assessment of functional and cognitive decline in older adults. However, understanding everyday activities is challenging: important information may come not only from conventional video, but also from human motion, depth, sound, object interactions, or different viewpoints. At the same time, continuously collecting all of this information may be impractical and can raise important privacy concerns.

This talk will introduce a new methodology for a seemingly fundamental problem. Given data (in any space, not necessarily real-valued) that are independent and identically distributed before and after a change point T, can we construct a confidence interval for T without any further assumptions? The answer is yes in both offline and sequential settings, and we will describe a universal method to achieve this goal. Experiments with images and text confirm that off-the-shelf pretrained classifiers, when combined with our methods, can yield sharp and valid confidence sets.

Structured data—spanning tables, time-series and relational databases —underpins high-impact predictive systems across retail, finance, healthcare, banking and core enterprise operations. Despite its ubiquity, standard predictive analytics workflows for structured data remain trapped in a bespoke and cumbersome train-per-dataset paradigm reliant on feature engineering, model selection, and hyperparameter tuning iterations. Over the past two years, large transformer-based pretrained foundation models for structured data have emerged to fundamentally challenge this status quo and offer state-of-the-art performance for predictive tasks like regression, classification and forecasting. By leveraging large-scale pretraining, these architectures enable cross-dataset transfer, handle heterogeneous schemas, and often outperform traditional supervised models, without the need for any dataset-specific training or tuning. In this talk, we explore the architectures, pretraining data and empirical capabilities of state-of-the-art predictive foundation models for structured data such as TimesFM and TabFM.

India stands at a defining moment in the use of artificial intelligence across healthcare and pharmaceuticals. An expanding digital health ecosystem, clinical data, advances in generative and agentic AI, and an innovation landscape are creating opportunities to improve access, affordability, quality, and speed of care. From AI-assisted screening, diagnostic imaging, and clinical decision support to virtual care and population health surveillance, intelligent technologies can extend expertise beyond urban centres. Across the pharmaceutical value chain, AI is accelerating target identification, drug discovery, clinical development, safety monitoring, manufacturing, supply chains, and evidence generation, while enabling more precise and personalised treatment. Yet transformation at scale requires more than high-performing algorithms. India's diversity of languages, populations, disease burdens, and care settings demands solutions that are locally validated, inclusive, and integrated into clinical workflows. Fragmented data, data quality, limited interoperability, algorithmic bias, privacy concerns, unclear accountability, and uneven digital readiness remain barriers. In high-stakes healthcare environments, trust must be earned through transparency, explainability, rigorous evaluation, human oversight, and continuous monitoring. This talk will examine the state of AI adoption in Indian healthcare and pharmaceuticals, highlight applications and opportunities, and explore the shift from predictive models to generative, multimodal, and agentic systems. It will also discuss the foundations needed for responsible scale: interoperable digital infrastructure, representative datasets, governance, multidisciplinary collaboration, and workforce capability. Ultimately, the future of AI in Indian healthcare will be defined not by technological sophistication alone, but by whether it delivers measurable, equitable, and trustworthy improvements for patients, clinicians, researchers, and the health system.

The pharmaceutical industry is under constant pressure to accelerate drug development while maintaining the highest standards of quality, compliance, and patient safety. Statistical programming plays a critical role in transforming clinical trial data into regulatory-ready analyses, tables, listings, figures, and submission deliverables. However, traditional programming workflows often involve significant manual effort in code development, validation, documentation, issue resolution, and reporting activities, contributing to extended development timelines. Recent advances in Generative AI present an opportunity to modernize statistical programming and clinical reporting processes. By using large language models, AI-powered coding assistants, and intelligent automation frameworks, organizations can streamline activities such as metadata interpretation, code generation, code review, test case creation, validation support, and automated report generation. This presentation explores practical applications of AI across the statistical programming lifecycle and discusses how AI-assisted workflows can improve efficiency while maintaining regulatory compliance and quality expectations. The session will highlight real-world opportunities, challenges, and future directions for AI-enabled clinical reporting.