Exp: Use Muti-objective learning by stevengogogo · Pull Request #26 · liruipeng/ml2

stevengogogo · 2025-06-17T19:20:51Z

This implementation provides an API for multi-objective learning.

where A is an aggregator that combines gradients

For $loss = 1.0 loss_{drm} + 500 loss_{bc}$

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 1  --bc_weight 1 --nx 2000  --aggregator "Constant" 1. 500.

For Dual Cone optimization

 python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 0  --bc_weight 1 --nx 2000  --aggregator "UPGrad"

This optimize pinn loss and boundary loss at the same time.

Monitor weights

add --monitor_aggregator

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 1  --bc_weight 1 --nx 2000  --gamma 0  --aggregator "NashMTL" 2 10. --monitor_aggregator

Weights: tensor([2.0694e-02, 3.3308e+01], device='cuda:0')
Cosine similarity: 0.9834

TorchJD package for Jacobian method

Many ways to combine gradients:

Choose maximium one (PINN loss and DRM loss help getting rid of local minima by gradient selection #25 )

stevengogogo · 2025-06-17T19:41:33Z

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-3 --activation tanh --sweeps 1 --hidden_dims 128 128 128 128 --high_freq 3 --loss_type 2 --bc_weight 1 --nx 2000  --gamma 0 --aggregator "None" --enforce_bc --plot

Training Level 0
Level 0: TRAIN
Iteration    499/ 10000, PINN+DRM Loss: -4.5650e+01, Err 2-norm:  1.3520e+01, inf-norm: 1.2588e+00
Iteration    999/ 10000, PINN+DRM Loss: -4.5722e+01, Err 2-norm:  1.3518e+01, inf-norm: 1.2586e+00
Iteration   1499/ 10000, PINN+DRM Loss: -4.5728e+01, Err 2-norm:  1.3517e+01, inf-norm: 1.2585e+00
Iteration   1999/ 10000, PINN+DRM Loss: -4.5465e+01, Err 2-norm:  1.3456e+01, inf-norm: 1.2543e+00
Iteration   2499/ 10000, PINN+DRM Loss: -4.5679e+01, Err 2-norm:  1.3579e+01, inf-norm: 1.2660e+00
Iteration   2999/ 10000, PINN+DRM Loss: -4.5720e+01, Err 2-norm:  1.3484e+01, inf-norm: 1.2543e+00
Iteration   3499/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3517e+01, inf-norm: 1.2585e+00
Iteration   3999/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3513e+01, inf-norm: 1.2583e+00
Iteration   4499/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3516e+01, inf-norm: 1.2584e+00
Iteration   4999/ 10000, PINN+DRM Loss: -4.5239e+01, Err 2-norm:  1.3345e+01, inf-norm: 1.2448e+00
Iteration   5499/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3519e+01, inf-norm: 1.2589e+00
Iteration   5999/ 10000, PINN+DRM Loss: -4.5709e+01, Err 2-norm:  1.3428e+01, inf-norm: 1.2505e+00
Iteration   6499/ 10000, PINN+DRM Loss: -4.5712e+01, Err 2-norm:  1.3508e+01, inf-norm: 1.2570e+00
Iteration   6999/ 10000, PINN+DRM Loss: -4.5659e+01, Err 2-norm:  1.3449e+01, inf-norm: 1.2502e+00
Iteration   7499/ 10000, PINN+DRM Loss: -4.5430e+01, Err 2-norm:  1.3549e+01, inf-norm: 1.2660e+00
Iteration   7999/ 10000, PINN+DRM Loss: -4.5727e+01, Err 2-norm:  1.3490e+01, inf-norm: 1.2558e+00
Iteration   8499/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3515e+01, inf-norm: 1.2583e+00
Iteration   8999/ 10000, PINN+DRM Loss: -4.5722e+01, Err 2-norm:  1.3495e+01, inf-norm: 1.2565e+00
Iteration   9499/ 10000, PINN+DRM Loss: -4.5729e+01, Err 2-norm:  1.3517e+01, inf-norm: 1.2583e+00
Iteration   9999/ 10000, PINN+DRM Loss: -4.5598e+01, Err 2-norm:  1.3552e+01, inf-norm: 1.2594e+00

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-3 --activation tanh --sweeps 1 --hidden_dims 128 128 128 128 --high_freq 3 --loss_type 2 --bc_weight 1 --nx 2000  --gamma 0 --aggregator "Random" --enforce_bc --plot

Iteration    499/ 10000, PINN+DRM Loss: 3.9525e+00, Err 2-norm:  1.2545e+01, inf-norm: 1.2468e+00
Iteration    999/ 10000, PINN+DRM Loss: -3.7693e+01, Err 2-norm:  1.1308e+01, inf-norm: 1.0476e+00
Iteration   1499/ 10000, PINN+DRM Loss: -4.2093e+01, Err 2-norm:  1.2699e+01, inf-norm: 1.2083e+00
Iteration   1999/ 10000, PINN+DRM Loss: -9.0663e+00, Err 2-norm:  1.2384e+01, inf-norm: 1.1649e+00
Iteration   2499/ 10000, PINN+DRM Loss: -3.1103e+01, Err 2-norm:  1.0250e+01, inf-norm: 9.4714e-01
Iteration   2999/ 10000, PINN+DRM Loss: -1.2633e+01, Err 2-norm:  1.3792e+01, inf-norm: 1.3089e+00
Iteration   3499/ 10000, PINN+DRM Loss: -4.0297e+01, Err 2-norm:  1.3636e+01, inf-norm: 1.2782e+00
Iteration   3999/ 10000, PINN+DRM Loss: -4.2600e+01, Err 2-norm:  1.4068e+01, inf-norm: 1.2996e+00
Iteration   4499/ 10000, PINN+DRM Loss: -3.1732e+01, Err 2-norm:  1.2627e+01, inf-norm: 1.1623e+00
Iteration   4999/ 10000, PINN+DRM Loss: -3.5047e+01, Err 2-norm:  1.2027e+01, inf-norm: 1.1159e+00
Iteration   5499/ 10000, PINN+DRM Loss: -4.2588e+01, Err 2-norm:  1.4576e+01, inf-norm: 1.3555e+00
Iteration   5999/ 10000, PINN+DRM Loss: -3.0203e+01, Err 2-norm:  1.3808e+01, inf-norm: 1.3062e+00
Iteration   6499/ 10000, PINN+DRM Loss: -3.2692e+01, Err 2-norm:  1.5385e+01, inf-norm: 1.4591e+00
Iteration   6999/ 10000, PINN+DRM Loss: -3.7795e+01, Err 2-norm:  1.1694e+01, inf-norm: 1.0892e+00
Iteration   7499/ 10000, PINN+DRM Loss: -3.8121e+01, Err 2-norm:  1.3208e+01, inf-norm: 1.2493e+00
Iteration   7999/ 10000, PINN+DRM Loss: -3.5225e+01, Err 2-norm:  1.4057e+01, inf-norm: 1.2916e+00
Iteration   8499/ 10000, PINN+DRM Loss: -3.6439e+01, Err 2-norm:  1.3031e+01, inf-norm: 1.1969e+00
Iteration   8999/ 10000, PINN+DRM Loss: -4.2368e+01, Err 2-norm:  1.2165e+01, inf-norm: 1.1320e+00
Iteration   9499/ 10000, PINN+DRM Loss: -3.7700e+01, Err 2-norm:  1.3979e+01, inf-norm: 1.3268e+00
Iteration   9999/ 10000, PINN+DRM Loss: -4.4185e+01, Err 2-norm:  1.3951e+01, inf-norm: 1.2921e+00

stevengogogo · 2025-06-17T23:02:19Z

Experiment Dual Cone method on DRM

Use $loss = loss_{drm} + 500 loss_{bc}$

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 1  --bc_weight 1 --nx 2000  --gamma 0  --aggregator "Constant" 1. 500. --plot

Iteration    999/ 10000, DRM Loss: -3.5050e+01, Err 2-norm:  8.1668e-01, inf-norm: 1.1278e-01

Dual Cone method for PINN + DRM

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 2  --bc_weight 1 --nx 2000  --gamma 0  --aggregator "UPGrad" --plot

Training Level 0
Level 0: TRAIN
Iteration    499/ 10000, PINN+DRM Loss: 7.7932e+00, Err 2-norm:  2.2756e+02, inf-norm: 1.6297e+01
Iteration    999/ 10000, PINN+DRM Loss: -3.0725e-01, Err 2-norm:  2.2581e+02, inf-norm: 1.6224e+01
Iteration   1499/ 10000, PINN+DRM Loss: -3.5438e+00, Err 2-norm:  2.2444e+02, inf-norm: 1.6135e+01
Iteration   1999/ 10000, PINN+DRM Loss: -4.0919e+00, Err 2-norm:  2.2425e+02, inf-norm: 1.6124e+01
Iteration   2499/ 10000, PINN+DRM Loss: -4.4187e+00, Err 2-norm:  2.2413e+02, inf-norm: 1.6117e+01
Iteration   2999/ 10000, PINN+DRM Loss: -4.7988e+00, Err 2-norm:  2.2398e+02, inf-norm: 1.6107e+01
Iteration   3499/ 10000, PINN+DRM Loss: -4.9994e+00, Err 2-norm:  2.2388e+02, inf-norm: 1.6102e+01
Iteration   3999/ 10000, PINN+DRM Loss: -5.0996e+00, Err 2-norm:  2.2383e+02, inf-norm: 1.6099e+01
Iteration   4499/ 10000, PINN+DRM Loss: -5.1486e+00, Err 2-norm:  2.2383e+02, inf-norm: 1.6096e+01
Iteration   4999/ 10000, PINN+DRM Loss: -5.1739e+00, Err 2-norm:  2.2382e+02, inf-norm: 1.6095e+01
Iteration   5499/ 10000, PINN+DRM Loss: -5.1908e+00, Err 2-norm:  2.2381e+02, inf-norm: 1.6095e+01
Iteration   5999/ 10000, PINN+DRM Loss: -5.2032e+00, Err 2-norm:  2.2381e+02, inf-norm: 1.6095e+01
Iteration   6499/ 10000, PINN+DRM Loss: -5.2282e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6096e+01
Iteration   6999/ 10000, PINN+DRM Loss: -5.2406e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6096e+01
Iteration   7499/ 10000, PINN+DRM Loss: -5.2289e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6094e+01
Iteration   7999/ 10000, PINN+DRM Loss: -5.2364e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6093e+01
Iteration   8499/ 10000, PINN+DRM Loss: -5.2364e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6093e+01
Iteration   8999/ 10000, PINN+DRM Loss: -5.2717e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6096e+01
Iteration   9499/ 10000, PINN+DRM Loss: -5.2721e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6095e+01
Iteration   9999/ 10000, PINN+DRM Loss: -5.2785e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6095e+01

Dual Cone method for PINN + DRM

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 2  --bc_weight 1 --nx 2000  --gamma 0  --aggregator "UPGrad" --plot

Training Level 0
Level 0: TRAIN
Iteration    499/ 10000, PINN+DRM Loss: 7.7932e+00, Err 2-norm:  2.2756e+02, inf-norm: 1.6297e+01
Iteration    999/ 10000, PINN+DRM Loss: -3.0725e-01, Err 2-norm:  2.2581e+02, inf-norm: 1.6224e+01
Iteration   1499/ 10000, PINN+DRM Loss: -3.5438e+00, Err 2-norm:  2.2444e+02, inf-norm: 1.6135e+01
Iteration   1999/ 10000, PINN+DRM Loss: -4.0919e+00, Err 2-norm:  2.2425e+02, inf-norm: 1.6124e+01
Iteration   2499/ 10000, PINN+DRM Loss: -4.4187e+00, Err 2-norm:  2.2413e+02, inf-norm: 1.6117e+01
Iteration   2999/ 10000, PINN+DRM Loss: -4.7988e+00, Err 2-norm:  2.2398e+02, inf-norm: 1.6107e+01
Iteration   3499/ 10000, PINN+DRM Loss: -4.9994e+00, Err 2-norm:  2.2388e+02, inf-norm: 1.6102e+01
Iteration   3999/ 10000, PINN+DRM Loss: -5.0996e+00, Err 2-norm:  2.2383e+02, inf-norm: 1.6099e+01
Iteration   4499/ 10000, PINN+DRM Loss: -5.1486e+00, Err 2-norm:  2.2383e+02, inf-norm: 1.6096e+01
Iteration   4999/ 10000, PINN+DRM Loss: -5.1739e+00, Err 2-norm:  2.2382e+02, inf-norm: 1.6095e+01
Iteration   5499/ 10000, PINN+DRM Loss: -5.1908e+00, Err 2-norm:  2.2381e+02, inf-norm: 1.6095e+01
Iteration   5999/ 10000, PINN+DRM Loss: -5.2032e+00, Err 2-norm:  2.2381e+02, inf-norm: 1.6095e+01
Iteration   6499/ 10000, PINN+DRM Loss: -5.2282e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6096e+01
Iteration   6999/ 10000, PINN+DRM Loss: -5.2406e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6096e+01
Iteration   7499/ 10000, PINN+DRM Loss: -5.2289e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6094e+01
Iteration   7999/ 10000, PINN+DRM Loss: -5.2364e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6093e+01
Iteration   8499/ 10000, PINN+DRM Loss: -5.2364e+00, Err 2-norm:  2.2379e+02, inf-norm: 1.6093e+01
Iteration   8999/ 10000, PINN+DRM Loss: -5.2717e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6096e+01
Iteration   9499/ 10000, PINN+DRM Loss: -5.2721e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6095e+01
Iteration   9999/ 10000, PINN+DRM Loss: -5.2785e+00, Err 2-norm:  2.2378e+02, inf-norm: 1.6095e+01

stevengogogo · 2025-06-17T23:41:08Z

Exp: Verify dual-cone optimization helps PINNs + soft BC loss

Dual Cone

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 0  --bc_weight 1 --nx 2000  --gamma 0  --aggregator "UPGrad" --plot

Training Level 0
Level 0: TRAIN
Iteration    499/ 10000, PINN Loss: 1.1153e+01, Err 2-norm:  1.4233e+01, inf-norm: 1.4310e+00
Iteration    999/ 10000, PINN Loss: 1.0686e+00, Err 2-norm:  5.0536e-01, inf-norm: 4.5360e-02
Iteration   1499/ 10000, PINN Loss: 3.5229e-01, Err 2-norm:  2.2734e-01, inf-norm: 2.2083e-02
Iteration   1999/ 10000, PINN Loss: 1.1591e-01, Err 2-norm:  1.0089e-01, inf-norm: 1.1598e-02
Iteration   2499/ 10000, PINN Loss: 5.2839e-02, Err 2-norm:  4.8972e-02, inf-norm: 5.6659e-03
Iteration   2999/ 10000, PINN Loss: 2.9012e-02, Err 2-norm:  3.4169e-02, inf-norm: 3.8587e-03
Iteration   3499/ 10000, PINN Loss: 1.9204e-02, Err 2-norm:  2.0535e-02, inf-norm: 2.3413e-03
Iteration   3999/ 10000, PINN Loss: 1.3088e-02, Err 2-norm:  1.8215e-02, inf-norm: 2.0128e-03
Iteration   4499/ 10000, PINN Loss: 1.4971e-01, Err 2-norm:  5.1648e-02, inf-norm: 7.5520e-03
Iteration   4999/ 10000, PINN Loss: 7.1656e-03, Err 2-norm:  1.2028e-02, inf-norm: 1.9021e-03
Iteration   5499/ 10000, PINN Loss: 4.7366e-03, Err 2-norm:  7.8400e-03, inf-norm: 9.1186e-04
Iteration   5999/ 10000, PINN Loss: 3.6847e-03, Err 2-norm:  6.0205e-03, inf-norm: 7.1356e-04
Iteration   6499/ 10000, PINN Loss: 2.9949e-03, Err 2-norm:  4.2069e-03, inf-norm: 3.9807e-04
Iteration   6999/ 10000, PINN Loss: 2.4976e-03, Err 2-norm:  3.9032e-03, inf-norm: 4.7027e-04
Iteration   7499/ 10000, PINN Loss: 2.1062e-03, Err 2-norm:  3.2904e-03, inf-norm: 3.9752e-04
Iteration   7999/ 10000, PINN Loss: 1.7800e-03, Err 2-norm:  2.6893e-03, inf-norm: 3.2555e-04
Iteration   8499/ 10000, PINN Loss: 1.5204e-03, Err 2-norm:  2.2626e-03, inf-norm: 2.7375e-04
Iteration   8999/ 10000, PINN Loss: 1.3033e-03, Err 2-norm:  1.9027e-03, inf-norm: 2.2980e-04
Iteration   9499/ 10000, PINN Loss: 3.6962e-02, Err 2-norm:  5.8394e-02, inf-norm: 7.6404e-03
Iteration   9999/ 10000, PINN Loss: 9.7147e-04, Err 2-norm:  1.1256e-03, inf-norm: 1.2255e-04

snapshot of weights of gradients

Weights: tensor([0.5813, 1.0964], device='cuda:0')
Cosine similarity: 0.9370
Weights: tensor([0.5808, 1.0967], device='cuda:0')
Cosine similarity: 0.9372
Weights: tensor([0.5804, 1.0970], device='cuda:0')
Cosine similarity: 0.9374

PINN with $loss = loss_{pinn} + loss_{bc}$

python pinn_1d.py   --levels 1 --epochs 10000 --lr 1e-4 --activation gelu --sweeps 1 --hidden_dims 256 256 256 --high_freq 3 --loss_type 0  --bc_weight 1 --nx 2000  --gamma 0 --plot

Iteration    499/ 10000, PINN Loss: 1.7111e+01, Err 2-norm:  3.9158e+00, inf-norm: 4.6693e-01
Iteration    999/ 10000, PINN Loss: 6.1949e-01, Err 2-norm:  1.1616e+00, inf-norm: 1.3581e-01
Iteration   1499/ 10000, PINN Loss: 1.9982e-01, Err 2-norm:  2.5898e-01, inf-norm: 3.0343e-02
Iteration   1999/ 10000, PINN Loss: 1.2159e-01, Err 2-norm:  1.2659e-01, inf-norm: 1.5461e-02
Iteration   2499/ 10000, PINN Loss: 7.9549e-02, Err 2-norm:  8.1205e-02, inf-norm: 9.7390e-03
Iteration   2999/ 10000, PINN Loss: 4.8445e-02, Err 2-norm:  4.6060e-02, inf-norm: 5.8364e-03
Iteration   3499/ 10000, PINN Loss: 3.3523e-02, Err 2-norm:  3.5849e-02, inf-norm: 4.3601e-03
Iteration   3999/ 10000, PINN Loss: 2.5072e-02, Err 2-norm:  2.8865e-02, inf-norm: 3.5170e-03
Iteration   4499/ 10000, PINN Loss: 1.9703e-02, Err 2-norm:  2.3881e-02, inf-norm: 2.9146e-03
Iteration   4999/ 10000, PINN Loss: 1.5842e-02, Err 2-norm:  2.0890e-02, inf-norm: 2.5257e-03
Iteration   5499/ 10000, PINN Loss: 1.3097e-02, Err 2-norm:  1.8221e-02, inf-norm: 2.1839e-03
Iteration   5999/ 10000, PINN Loss: 1.1010e-02, Err 2-norm:  1.6584e-02, inf-norm: 1.8699e-03
Iteration   6499/ 10000, PINN Loss: 9.4804e-03, Err 2-norm:  1.4153e-02, inf-norm: 1.6923e-03
Iteration   6999/ 10000, PINN Loss: 8.2994e-03, Err 2-norm:  1.2402e-02, inf-norm: 1.5738e-03
Iteration   7499/ 10000, PINN Loss: 2.5371e-02, Err 2-norm:  3.7283e-02, inf-norm: 3.5382e-03
Iteration   7999/ 10000, PINN Loss: 9.1092e-03, Err 2-norm:  1.5372e-02, inf-norm: 2.2640e-03
Iteration   8499/ 10000, PINN Loss: 5.9730e-03, Err 2-norm:  8.5018e-03, inf-norm: 1.0841e-03
Iteration   8999/ 10000, PINN Loss: 5.4161e-03, Err 2-norm:  8.3174e-03, inf-norm: 9.8834e-04
Iteration   9499/ 10000, PINN Loss: 4.9422e-03, Err 2-norm:  7.5278e-03, inf-norm: 8.9516e-04
Iteration   9999/ 10000, PINN Loss: 2.7889e-02, Err 2-norm:  4.5591e-02, inf-norm: 4.5103e-03

stevengogogo · 2025-06-18T17:23:06Z

@liruipeng Please take a look. It's ready. The main change is the loss function now has two output loss, losses:list. This allow multi-objective learning possible and allow for printing out all losses separatedly

stevengogogo · 2025-06-20T18:58:19Z

Fix CI package and conflict

stevengogogo · 2025-06-27T17:19:52Z

        u = model.get_solution(x)
        loss = loss_func(u, mesh.u_ex)
-        return loss
+        return loss, [loss,]


major change

stevengogogo added 4 commits June 17, 2025 13:23

explicit return loss terms

e8aa25b

add torchjd for multi task learning

69fe618

add aggragator

86a71ac

fix call agg api

6eb1b92

stevengogogo added the experiment label Jun 17, 2025

stevengogogo linked an issue Jun 17, 2025 that may be closed by this pull request

PINN loss and DRM loss help getting rid of local minima by gradient selection #25

Open

add monitor_aggregator

18d89a9

stevengogogo added 2 commits June 17, 2025 18:28

allow strin got int or float

3eb02eb

ignore launch.json

839071a

Merge branch 'main' into stc/add-gradient-selection

f3f4590

stevengogogo marked this pull request as ready for review June 18, 2025 17:22

stevengogogo requested a review from liruipeng June 18, 2025 17:22

stevengogogo added the enhancement New feature or request label Jun 18, 2025

stevengogogo mentioned this pull request Jun 18, 2025

Experiment on Dual cone optimizer #24

Draft

stevengogogo and others added 6 commits June 20, 2025 11:38

Merge branch 'main' into stc/add-gradient-selection

b69e748

add agg package

f63861c

update torchjd

ce08e60

fix indent issue

4f4ca96

add import monitor aggregator

7988125

fix merge bug: argument error

8004103

stevengogogo commented Jun 27, 2025

View reviewed changes

stevengogogo added the RFR ready for review label Oct 5, 2025

siuwuncheung self-requested a review October 10, 2025 17:52

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Exp: Use Muti-objective learning #26

Exp: Use Muti-objective learning #26
stevengogogo wants to merge 14 commits into
mainfrom
stc/add-gradient-selection

stevengogogo commented Jun 17, 2025 •

edited

Loading

Uh oh!

stevengogogo commented Jun 17, 2025 •

edited

Loading

Uh oh!

stevengogogo commented Jun 17, 2025 •

edited

Loading

Uh oh!

stevengogogo commented Jun 17, 2025 •

edited

Loading

Uh oh!

stevengogogo commented Jun 18, 2025 •

edited

Loading

Uh oh!

stevengogogo commented Jun 20, 2025 •

edited

Loading

Uh oh!

stevengogogo Jun 27, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

stevengogogo commented Jun 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

For $loss = 1.0 loss_{drm} + 500 loss_{bc}$

For Dual Cone optimization

Monitor weights

TorchJD package for Jacobian method

Uh oh!

stevengogogo commented Jun 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stevengogogo commented Jun 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Experiment Dual Cone method on DRM

Dual Cone method for PINN + DRM

Dual Cone method for PINN + DRM

Uh oh!

stevengogogo commented Jun 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Exp: Verify dual-cone optimization helps PINNs + soft BC loss

Dual Cone

PINN with $loss = loss_{pinn} + loss_{bc}$

Uh oh!

stevengogogo commented Jun 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stevengogogo commented Jun 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stevengogogo Jun 27, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

stevengogogo commented Jun 17, 2025 •

edited

Loading

stevengogogo commented Jun 17, 2025 •

edited

Loading

stevengogogo commented Jun 17, 2025 •

edited

Loading

stevengogogo commented Jun 17, 2025 •

edited

Loading

stevengogogo commented Jun 18, 2025 •

edited

Loading

stevengogogo commented Jun 20, 2025 •

edited

Loading