Conference Round-Up: CDC 2025 and NeurIPS 2025
Researchers across the Erlangen AI Hub continue to showcase their work on the international stage. This season, Hub members presented at the IEEE Conference on Decision and Control (CDC 2025) and NeurIPS 2025, one of the world’s leading AI gatherings. Their contributions span advances in autonomous systems, the mathematical foundations of control, and the growing use of generative AI in finance. The highlights are captured below.
Advances in Abstraction-Based Control at CDC 2025
Designing safe, reliable controllers for autonomous systems, from drones to self-driving vehicles, remains a fundamental challenge in AI. At CDC 2025, Erlangen Hub PDRA Thom Badings and Co-Investigator Alessandro Abate presented new research advancing abstraction-based control, a principled approach for computing correct-by-construction control policies under uncertainty.
Their two papers deliver key contributions:
- Strengthening the mathematical foundations
A refined abstraction framework capable of computing provably safe control policies even when system dynamics are uncertain. This work enhances both precision and scalability for complex autonomous platforms. - Introducing data-driven abstraction methods
New techniques for constructing abstractions directly from empirical data, reducing reliance on fully specified analytical models and enabling robust control in partially known environments.
These developments push forward the frontier of reliable autonomous decision-making and contribute to the Hub’s broader mission to develop rigorous foundations for trustworthy AI.
Further reading:
Probabilistic Alternating Simulations for Policy Synthesis in Uncertain Stochastic Dynamical Systems https://arxiv.org/abs/2508.05062
• Data-Driven Abstraction and Synthesis for Stochastic Systems with Unknown Dynamics: https://arxiv.org/abs/2508.15543
Generative AI for Finance: Rama Cont at NeurIPS 2025
At the NeurIPS 2025 Workshop on Generative AI in Finance, Erlangen Hub Co-Investigator Rama Cont delivered an invited talk on how generative models are transforming quantitative finance.
Financial markets are noisy, nonlinear, and highly interdependent, making simulation and risk assessment especially challenging. Cont presented recent work demonstrating how GAN-based models can emulate complex market behaviour, generate realistic scenarios, and support robust risk management.
His talk covered several key generative approaches developed by Cont and collaborators, including:
- VolGAN for stochastic volatility surfaces
- Tail-GAN for modelling rare but high-impact tail events
- YieldGAN for yield curve dynamics
- Data-driven hedging with generative models, a method using conditional generative models to compute hedge ratios across simulated market scenarios
The last of these was the focus of his presentation and recent paper, which proposes a non-parametric approach to hedging that outperforms classical delta and delta-vega strategies, even years after the training period.
The workshop itself featured leading voices from academia and industry, reflecting the rapid growth of AI-driven approaches in financial modelling.
Paper abstract:
Cont, R., Vuletić, M. Data-driven hedging with generative models. Ann Oper Res (2025)
Hub Leadership at NeurIPS 2025
NeurIPS 2025 was among the most competitive editions of the conference to date, with just 24.5% of submissions accepted. Against this backdrop, Erlangen AI Hub Director Michael Bronstein appeared as a co-author on nine accepted papers, presented across poster and spotlight sessions.
These contributions span generative and diffusion models, flow-based methods, equivariant and graph neural architectures, optimisation, and inference. All are core areas in the mathematical foundations of modern AI, and together, they reflect sustained engagement with both the theory and practice of scalable learning systems.
In a conference landscape increasingly shaped by large North American corporations and Chinese research institutions, this level of representation places Bronstein as a key figure in small group of Europe-based researchers maintaining strong technical visibility at NeurIPS, while highlighting the continued contribution of UK and European research to foundational questions shaping the field.
Further reading