Conditional Flow Matching for Engineering Inverse Design
Models, datasets, and reproducibility artifacts for the EngOpt 2026 conditional flow-matching paper.
UpdatedNote Primary contribution: selected CFM checkpoints for 10 seeds, two EngiBench tasks, and the evaluated solver/NFE configurations. Beams2D MMD/DPP metadata records the sigma=1 correction. W&B report: https://wandb.ai/smassoudi-eth-z-rich/engiopt-flow-matching/reports/Conditional-Flow-Matching-for-ML-Based-Inverse-Design-Problems---Experiments-and-Reproducibility--VmlldzoxNzgzMTUyMQ==
IDEALLab/engiopt-engopt2026-diffusion-2d-cond
UpdatedNote Corrected EngiOpt conditional diffusion baseline with 1000 denoising steps; one selected checkpoint per seed and task.
IDEALLab/engiopt-engopt2026-cgan-cnn-2d
UpdatedNote Corrected EngiOpt cGAN baseline with sigmoid generator output; one selected checkpoint per seed and task.
IDEALLab/beams_2d_50_100_v0
Viewer • Updated • 4.85k • 2.05kNote EngiBench Beams2D dataset used for the 100 x 50 structural-design experiments.
IDEALLab/heat_conduction_2d_v0
Viewer • Updated • 441 • 958Note EngiBench HeatConduction2D dataset used for the thermal-design experiments.