Tadpole (NeurIPS 2026)
github.com/tum-pbs/tadpoleAutoencoder-based foundation model for 3D PDEs pretrained with online learning: a GPU-based synthetic-data solver generates training samples on the fly (no storage/IO bottleneck), scaling to an equivalent of hundreds of terabytes; learns transferable representations across heterogeneous physical systems with varying state variables and resolutions, and supports autoencoding, dynamics prediction via parameter-efficient LoRA fine-tuning, and generative modeling, with pretrained weights on HuggingFace (TU Munich Thuerey Group, 29+ stars, Apache 2.0, 2026)
Sourced from
- Awesome AI for Science — github.com/tum-pbs/tadpole
Related resources
Kolmogorov-Arnold Networks with learnable activation functions on edges instead of fixed node activations, achieving strong performance in function fitting, PDE solving, and scientific discovery with enhanced interpretability as an alternative to MLPs (MIT, 16.3K+ stars, 2024)
High-performance symbolic regression for discovering interpretable scientific equations from data, multi-population evolutionary search with Python/Julia backend, widely used in physics and astronomy (Cambridge, NeurIPS 2023)
Learning operators in Fourier space
Differentiable PDE solving framework for machine learning with built-in fluid simulation, supporting PyTorch/JAX/TensorFlow backends and enabling neural network training within physical simulations (TUM, MIT License)
Sparse identification of nonlinear dynamics
Python package for Dynamic Mode Decomposition (DMD) and its variants (exact, incremental, multiresolution MR-DMD, higher-order HODMD, parametric, BOP-DMD, DMD-control, and DL-DMD neural-network extensions), enabling data-driven reduced-order modeling, Koopman operator analysis, and flow-field feature extraction for high-dimensional dynamical systems in fluid dynamics and scientific computing (1.2K+ stars, MIT License, actively maintained)