ABSTRACT

Artificial intelligence is rapidly transforming scientific research, industry, and society. Among its recent advances, generative models have attracted particular attention for their ability to produce text, images, and complex structured outputs. Despite their computational sophistication, these systems remain deeply connected to core statistical ideas.

This keynote speech explores the evolving relationship between statistics and artificial intelligence, highlighting how probabilistic modelling, inference, and learning form a common foundation for both classical statistical methods and modern AI architectures. Rather than presenting these fields as separate or competing domains, the talk emphasizes their conceptual continuity and mutual reinforcement.

The lecture will reflect on how statistical thinking contributes to reliability, uncertainty quantification, interpretability, and responsible deployment of generative systems. By viewing artificial intelligence through a statistical lens, we gain a clearer understanding of both its power and its limitations, and we identify promising directions for future research at the intersection of statistics, mathematics, and computational science.

ABSTRACT

Statistical education in the Philippines stands at a critical intersection of academic development, governance reform, and national capacity building. In an era increasingly shaped by data-driven decision-making, the ability to produce, analyze, and interpret reliable statistics is fundamental to evidence-based policy, democratic accountability, and sustainable development. Yet foundational challenges persist. Recent international assessments, such as PISA 2022, reveal significant gaps in mathematical proficiency, constraining the pipeline of students prepared for advanced quantitative training. At the tertiary level, while the Commission on Higher Education has institutionalized competency-based standards for the Bachelor of Science in Statistics, enrollment in specialized statistics programs remains limited relative to growing national demand.

This plenary examines the state of statistical education within the broader institutional framework of the Philippine Statistical System. It situates higher education within the mandates of Republic Act No. 10625 (Philippine Statistical Act of 2013), which designates the Philippine Statistical Research and Training Institute (PSRTI) as the research and training arm of the system, and the Philippine Statistical Development Program (PSDP) 2023–2029, which prioritizes modernization and human capital development. The discussion also highlights the Community-Based Monitoring System Act (RA 11315), emphasizing the localization of statistical capacity and PSRTI’s role in strengthening technical competencies at the local government level.

By integrating educational, institutional, and policy perspectives, the plenary underscores that statistical education is not merely an academic undertaking but a strategic national investment. Strengthening foundational numeracy, modernizing curricula, fostering research alignment with national priorities, and deepening collaboration between academe and the official statistical system are essential steps toward building a resilient, future-ready statistical workforce for the Philippines.

Insert Abstract Here

The independence test among multivariate continuous vectors is a fundamental but complex task in statistics. Our contribution lies in the development of an L1 -norm--based test designed to assess multivariate independence, thus representing a first of its kind. Although the L1 -based test---defined between the joint density function and the product of the marginal densities associated with the presumed independent vectors---has attractive features, its theoretical foundations have been difficult to establish. Under the null hypothesis, we use Poissonization techniques to derive the asymptotic normal approximation of the corresponding test statistic, without imposing any regularity assumptions on the underlying Lebesgue density function. Remarkably, we find that the limiting distribution of the L1 -based statistics does not depend on the model. This unexpected result contributes to the robustness and versatility of our method. Moreover, our tests exhibit nontrivial local power against a subset of local alternatives that converge to the null hypothesis at the rate

 n-1/2hn-d/4

where $n$ denotes the sample size and the hn bandwidth. Finally, the theory is supported by an extensive simulation study aimed at examining the finite-sample performance of our proposed test. The results show that our testing procedure generally outperforms existing approaches across the various scenarios considered. Joint work with N. Berrahou and L. Douge. 

ABSTRACT

The growing availability of data and computational tools has transformed how statistical models are applied in real-world decision making, particularly in the insurance and financial sectors. This talk presents several research projects arising from collaborations between the Department of Mathematics, Universitas Gadjah Mada, and industry partners in the insurance sector. The studies illustrate how statistical modelling and data-driven predictive analytics can be translated into practical decision-support systems for industry applications, including insurance pricing using Bayesian statistical models, portfolio optimization under regulatory constraints, and the development of predictive analytics tools for financial reporting under the IFRS-17 standard. These projects demonstrate how mathematical and statistical research can be integrated with industry data and operational systems to support risk analysis, financial decision making, and regulatory compliance. The collaboration has produced scientific publications, intellectual property outputs such as patents and copyrighted software, and practical analytical tools that contribute to improving efficiency and risk management in insurance companies.

ABSTRACT

Conditional distribution estimation (CDE) is a key tool in nonparametric forecasting and risk analysis. While important progress has been made in finite-dimensional and stationary settings, functional data and nonstationary dynamics introduce substantial additional challenges.

In this study, we propose a Nadaraya–Watson–type conditional quantile estimator for regularly mixing locally stationary functional time series. The approach combines three kernel components: a time-rescaling kernel to account for local stationarity, a functional kernel to measure similarity between infinite-dimensional covariates, and an integrated kernel corresponding to the cumulative distribution function of the response variable.

We discuss the theoretical consistency of the estimator and investigate its finite-sample performance through numerical experiments. The methodology is further illustrated with an application to financial data, specifically the Nikkei 225 index, highlighting its relevance for forecasting and risk assessment in evolving functional environments.

ABSTRACT

In this article, we present a stochastic model for the movement of Helium particles within a graphite channel, focusing on Knudsen diffusion. We develop a semi-Markov model to describe the movement of the particle, derive the stationary distribution of its mean position, and analyze the model’s asymptotic properties. To validate the model, we compare its theoretical outcomes with Monte Carlo simulations. As temperature significantly influences on the movement of particles, two situations are studied for high and low temperature. In both cases, theoretical and simulation results by Monte Carlo coincide. Furthermore, we propose estimation methods for the local parameters of the model and demonstrate its application using data from Molecular Dynamics simulations.