Levente Zólyomi

NXAI · Institute for Machine Learning JKU Linz. Advised by Prof. Sepp Hochreiter.

prof_pic.jpg

I’m a PhD Researcher at NXAI and a PhD student at the Institute for Machine Learning, JKU Linz, advised by Prof. Sepp Hochreiter. Before starting my PhD, I was a researcher at King AI Labs (Microsoft Gaming) in Stockholm, Sweden. I hold a joint master’s degree from Aalto University (Helsinki, Finland) and Eindhoven University of Technology (Eindhoven, Netherlands).

My research focuses on efficient subquadratic architectures for sequence modeling. In particular, I work on time series foundation models and linear and nonlinear RNNs as scalable alternatives to Transformers, as well as self-supervised learning methods for sequential data.

selected publications

  1. NeurIPS
    Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models
    Sofiane Ennadir*, Levente Zólyomi*, Oleg Smirnov, Tianze Wang, John Pertoft, Filip Cornell, and Lele Cao
    In NeurIPS 2025, 2025
    HTML PDF
  2. ICLR
    Understanding Early Collapse in Predictive World-Model Pretraining
    Sofiane Ennadir*, Levente Zólyomi*, and Oleg Smirnov*
    In ICLR 2026 Workshop on World Models, 2026
    HTML PDF
  3. ICML
    Nonlinear RNNs as a Compute Shortcut for Time Series Foundation Models
    Levente Zólyomi, David Stap, Sebastian Böck, Günter Klambauer, and Sepp Hochreiter
    In Foundational Models for Structured Data @ ICML 2026, 2026