Projects / Dynamical systems & time

Khresmoi: forecasting dynamical systems

Neural Koopman operators, HiPPO memories, synthetic gradients and symbolic dynamics, tested on synthetic oscillators and chaotic systems.

  • Learning Koopman Invariant Subspaces (LKIS) and the Koopman Neural Forecaster
  • HiPPO-LegS, including a transformer with a HiPPO encoder
  • Synthetic gradients computed with torch.func
  • A symbolic forecaster built on residual vector quantization
  • Contrastive learning with matrix-power layers
  • Test systems: FitzHugh–Nagumo, coupled FHN, and a smooth time-series generator