RESEARCH

I study continuous functions in physics and computational chemistry, with a current focus on continuous representations to model physical quantities such as DFT electron densities.

I’m also interested in representation learning: for materials and molecules, how can we represent atoms and structure more richly than categorical types, to enable zero- or few-shot generative modeling for material discovery?

Questions or collaborations? Email me via the About page.

Selected Publications

• CORDS: Continuous Representations of Discrete Structures
  Hadži Veljković et al., 2026 — Invertible mapping from variable-size sets to continuous density + feature fields, enabling models to operate in field space while decoding exactly back to discrete objects; evaluated on molecular generation, detection, and simulation-based inference.
• Dynamic Training Enhances Machine Learning Potentials for Long-Lasting Molecular Dynamics
  Žugec, Hadži Veljković et al., 2025 — Introduces dynamic training to refresh ML interatomic potentials during long MD runs, boosting EGNN accuracy on hydrogen–palladium/graphene systems and maintaining fidelity over extended simulations.
• Fast yet Safe: Early-Exiting with Risk Control
  Jazbec, Timans, Hadži Veljković et al., 2024 — Applies distribution-free risk control to early-exit neural nets so exits trigger only when predictions meet quality guarantees, delivering speedups without sacrificing target accuracy across vision and language tasks.