Dear Colleagues,

I'm pleased to invite you to the Department Seminars, which will be held on  Wednesday, 4th of March, 10:00-12.30 CET, in Meeting Room 1 with the following schedule:

- 10:00-11:00 CET: Rajarshi Guhaniyogi (Texas A&M University), Bridging Statistical, Scientific and Artificial Intelligence: Trustworthy Deep Neural Networks for Complex Structured Data

- 11.30- 12.30 CET: Sharmistha Guha (Texas A&M University), Neural Network Gaussian Processes for Multiplex Networks: Joint Modeling of Dynamics and Attributes under Partial Observation

All the interested people are warmly invited to attend in person or online!

Best regards
Roberto Casarin (and on behalf of the Organizing Committee)

More information available here:
Link Zoom: bit.ly/insem-2425
ID: 880 2639 9452
Passcode: InSem-2425
https://www.unive.it/data/agenda/3/110469
https://www.unive.it/data/agenda/3/110466


Title:
Bridging Statistical, Scientific and Artificial Intelligence: Trustworthy Deep Neural Networks for Complex Structured Data

Abstract:
The explosive expansion of large, structured datasets is radically transforming the landscape of statistical inference, unlocking unprecedented possibilities while simultaneously introducing formidable challenges. Although hierarchical Bayesian methods offer a gold standard for principled inference and rigorous uncertainty quantification, they falter in terms of scalability when confronted with the high dimensionality and sheer scale of modern data. Deep Neural Networks (DNNs) have made remarkable strides on the scalability front, yet their use in the literature remains largely as opaque, black-box that are ill-suited for inference, particularly with structured data. To overcome these barriers, we introduce DNN-based generative models, precisely engineered for two complex and impactful domains: (i) functional output regression with functional and network-valued inputs, and (ii) functional factor modeling for multivariate functional datasets. Our framework leverages the connection between variational deep Gaussian processes and DNNs, delivering transparent and interpretable inference via well-calibrated uncertainty quantification, and achieving rapid, Markov Chain Monte Carlo (MCMC)-free optimization for high-resolution, large-sample scenarios. We reveal how Bayesian nonparametric principles and cutting-edge deep learning can be seamlessly unified, positioning deep Gaussian process priors at the forefront for scalable generative modeling of object-valued data. Demonstrating the framework’s practical strength, we showcase compelling real-world applications, including image-on-image regression and large-scale remote sensing for the carbon-water cycle. This work is in collaboration with the postdoctoral scholar Dr. Yeseul Jeon, as well as scientists from the UC San Francisco Medical School.

Title:
Neural Network Gaussian Processes for Multiplex Networks: Joint Modeling of Dynamics and Attributes under Partial Observation

Abstract:
Terrorism networks are dynamic, multiplex, and often partially observed, demanding uncertainty-aware inference. This talk presents Dynamic Joint Learner, a Bayesian framework that jointly models the co-evolution of multiplex layers and node attributes using shared, time-varying latent factors. These latent trajectories are governed by neural network Gaussian processes, combining deep-network expressiveness with principled uncertainty propagation. The method supports predictive inference on hidden links, evolving organizational attributes (size, ideology, leadership, operational capacity), and emergent communities, including friend-foe structures. Simulation studies and an application to interactions among prominent terrorist organizations show improved performance over existing approaches for link prediction, attribute forecasting, and clustering, with calibrated uncertainty. The framework offers a practical toolkit for analysts working with partially observed, co-evolving networks and is broadly applicable beyond counter-terrorism.