IVADO Thematic Semester - Statistical Foundations of AI

Artificial Intelligence (AI) is increasingly shaping scientific inquiry, industry, and public life, yet its rapid development often obscures the statistical principles that underpin its methods. Organizing a thematic semester dedicated to the statistical foundations of AI offers a timely and rigorous opportunity to strengthen the conceptual and technical bridge between statistics and AI. Moreover, this focused initiative serves the dual purpose of reinforcing foundational research while equipping the statistical community to critically assess and innovate upon AI’s applications across diverse contexts. It addresses a central need: situating the power of AI firmly within the statistical reasoning that guarantees its reliability and trustworthiness.

The semester will open in May 2026 with a five-day bootcamp designed to equip statisticians with a working command of advanced AI tools, including large language models (LLMs), deep learning, reinforcement learning, agent-based models, and AI alignment frameworks. Participants will gain hands-on experience with modern AI pipelines while revisiting their statistical underpinnings - modeling assumptions, estimation principles, and sources of uncertainty.

Following the bootcamp, a five-day workshop on “Statistics in Trustworthy AI” will examine the statistical backbone of reliable, auditable, and fair artificial intelligence. As AI systems move into high-stakes domains such as health, finance, and public services, the need for evidence-based governance becomes pressing. Much of current AI practice is benchmark-driven and metric-optimized, yet real-world deployment demands error control, calibrated uncertainty, fairness under distribution shift, and auditability by regulators and clinicians. This workshop will bridge statistics and AI to address these deployment challenges, showing how statistical principles translate into trustworthy AI systems - not just in theory, but in tools ready for practical use.

In June 2026, a four-day workshop on “Uncertainty in AI” will explore how the statistical sciences are advancing our understanding of uncertainty in intelligent systems. Uncertainty arises in many forms - from the predictive accuracy of deep learning algorithms and hallucinations in generative AI to a still largely empirical understanding of what AI is, and is not, capable of doing. This workshop will highlight principled ways in which statistical theory and methods contribute to our evolving understanding of the uncertainty that accompanies AI. Topics will range from Bayesian deep learning to the analysis of transfer learning via optimal transport and flow matching, and from the mathematics of transformers and interacting particle systems to emerging formal characterizations of chain-of-thought reasoning.

The thematic semester will conclude in August 2026 with a two-day symposium, “AI Meets Statistics: Biomedical Data Perspectives,” dedicated to the interplay between AI and statistics in health-related research. This symposium will be co-organized with the Centre de Recherches Mathématiques (CRM) as part of its thematic semester “Math for Health,” to be held from August to December 2026. Biomedical research is being transformed by data of unprecedented scale and complexity - population-level genomics, high-resolution biological imaging, and expansive electronic health records (EHRs) are reshaping drug development and precision medicine. Yet current statistical and computational methods struggle to meet the challenges of heterogeneity, noise, high dimensionality, and regulatory rigor. This symposium will bring together statisticians, machine learning researchers, and biomedical scientists to build a shared methodological foundation for this emerging field. Core themes will include high-dimensional inference in genomics, representation learning for imaging, causal discovery from EHR data, and uncertainty quantification for decision-critical applications in drug development.

KEY DATES

1. Bootcamp:  Statistical Insights into Modern AI Systems - May 4th - May 8th, 2026

2. Workshop 1: Statistics in Trustworthy AI - May 11th - May 15th, 2026

3. Workshop 2: Uncertainty in AI - June 8th - June 11th, 2026

4. Symposium: AI Meets Statistics: Biomedical Data Perspectives - August 20th - August 21st, 2026