Oleg (Ole) Szehr,
AI Executive & Scientific Leader

AI executive and scientific leader with over a decade of experience spanning pure and applied mathematics, quantitative finance, and artificial intelligence.
I serve as Scientific Lead of the strategic AI collaboration between IDSIA and UBS AG, overseeing more than ten concurrent initiatives and ~100 contributors across research, engineering, and business units. I define research directions and system architectures, set strategic goals, and oversee the development of production-grade AI systems.
In my professional life, I have had the opportunity to work in top-tier programs and institutions — studying at ETH, earning scholarships for master's and doctoral studies, working as a mathematician at the DAMTP in Cambridge, and as a Quant at Credit Suisse and AcadiaSoft LSEG (formerly Quaternion).
Alongside industry-focused initiatives, my group pursues fundamental research in mathematical modeling and the theory of artificial intelligence, with particular emphasis on reinforcement learning and decision systems.

Contact information:

Oleg Szehr

Dalle Molle Institute for Artificial Intelligence USI-SUPSI
Polo universitario Lugano - Campus Est, Via la Santa 1
CH-6962 Lugano-Viganello
E-Mail: oleg.szehr@idsia.ch
IDSIA directions and contact information.

Research management

IDSIA has been selected as a strategic partner institution to conduct AI research at global scale for UBS. In this role my responsibilities include:

• Aligning IDSIA’s financial AI research with UBS’s strategic priorities, ensuring impactful innovation
• Leading and mentoring research teams to develop cutting-edge AI solutions for UBS
• Designing AI system architectures, selecting models and defining the business model specifications
• Recruiting and nurturing top-tier AI talent to strengthen our collaboration with UBS

Scientific Interests

My scientific interests lie at the intersection of mathematics, algorithms, and intelligent systems. I specialize in developing mathematically rigorous models and algorithms, and implementing them in industry systems. My research spans a broad range of topics in algorithmic mathematics — including quantum computing, kernel methods, and deep reinforcement learning — with a recurrent focus on robustness and performance guarantees. At present, my group concentrates on investigating deep reinforcement learning, both from a theoretical and applied perspective. More recently, I have assumed a scientific management role, overseeing interdisciplinary research efforts in AI.

Mathematics, Algorithms, Computation

In computational mathematics, I am interested in reproducing kernel methods for large (infinite-dimensional) systems. This interest originates from challenges in quantum computing, such as analyzing the convergence speed of quantum algorithms and ensuring the stability of quantum memories — domains where performance guarantees are essential, even under worst-case conditions. In the context of computational stability analysis, I have applied kernel methods to obtain estimates on resolvents, which play a key role in assessing the numerical stability of algorithms. A notable result from this line of work was resolving a long-standing open problem on the stability of operator inversion, where Banach space kernel methods provide explicit classes of worst-case operators. Much of this research was grounded in the asymptotic analysis and approximation of kernels, driven by the practical necessities of analyzing large-scale systems. More recently, I have been applying these techniques to kernel-based digital signal processing, with a focus on analyzing aliasing effects in increasingly high-frequency signals.

Reinforcement Learning, Financial Artificial Intelligence

In reinforcement learning and financial AI, my group investigates a broad range of topics, spanning from mathematical foundations to the development of industry systems. Continuing my line of work on convergence and stability, we recently analyzed the behavior of upside-down reinforcement learning, contributing new insights into its theoretical properties. We frequently employ game-theoretic frameworks to design algorithms for financial decision-making. On the applied side, recent research has focused on reframing financial planning problems — such as derivatives hedging and balance sheet management — as games, and solving them with modern reinforcement learning techniques.

Publications

A complete collection of my research works is available from Google Scholar, ArXiv and SSRN. A list of selected publications with descriptions is here.