Generative AI (GenAI) has seen tremendous advances driven by deep learning, evolving from early energy-based and latent variable models to the recent and more expressive frameworks such as score-based, diffusion, and flow-based generative models, among other cutting-edge paradigms. These models have demonstrated remarkable capabilities in learning complex data distributions and capturing underlying structures in high-dimensional spaces. Beyond conventional data synthesis, modern generative models offer probabilistic predictions and Bayesian interpretations, enabling uncertainty-aware and data-driven decision-making. Their probabilistic nature allows them to address a broader range of problems than pure data generation, supporting applications such as forecasting, inverse problem-solving, control, and scientific discovery, where modelling uncertainty, trustworthiness, and interpretability are crucial for diverse real-world domains. This workshop aims to explore advances in reliable generative modelling, including methods for uncertainty quantification, robustness under distributional shift or concept drift, and calibration of probabilistic outputs. We particularly encourage interdisciplinary contributions bridging deep generative modelling, neural network theory, and probabilistic inference, promoting innovative applications across vision, language, signal processing, robotics, and complex systems modeling. This workshop will focus on two main aspects:
Demonstrating the applicability of GenAI across diverse domains beyond traditional data generation and synthesis. We aim to showcase how modern generative models can be leveraged for decision-making, modeling complex systems, and real-world applications in areas such as scientific computing, robotics/autonomous vehicles, signal processing, health, and more.
Exploring the predictive capabilities of GenAI models, including uncertainty quantification and probabilistic reasoning. The workshop emphasizes how generative models can provide probabilistic predictions, estimate uncertainty, and support informed decision-making, highlighting their Bayesian or probabilistic foundations.
Together, these two themes reinforce the reliability of Generative AI, encouraging its confident use in real-world, safety-critical, and computer-aided applications. Topics of interest include, but are not limited to:
Generative Neural Networks;
Score-based, flow-based, and other recent generative methodologies;
Generative AI for trustworthy, explainable, and interpretable machine learning;
Generative AI for data-driven decision making;
Probabilistic and Bayesian generative modeling;
Uncertainty-aware reinforcement learning and generative control;
Application of generative AI beyond standard data synthesis;
Evaluation metrics, benchmarks, and reproducibility in generative AI;
Cross-disciplinary and real-world applications of generative models.