VITA-Group/Bi-RPT

[CPAL 2024] "Cross-Quality Few-Shot Transfer for Alloy Yield Strength Prediction: A New Materials Science Benchmark and A Sparsity-Oriented Optimization Framework" by Xuxi Chen, Tianlong Chen, Everardo Yeriel Olivares, Kate Elder, Scott K. McCall, Aurelien Pierre Philippe Perron, Joseph T. McKeown, Bhavya Kailkhura, Zhangyang Wang, Brian Gallagher

Cross-Quality Few-Shot Transfer for Alloy Yield Strength Prediction: A New Materials Science Benchmark and A Sparsity-Oriented Optimization Framework

Environment

conda env create -f environment.yml
conda activate birpt

Datasets

The dataset can be found in this link. It can be also found in this link.

Experiments

Proof-of-concept Experiments

Experiments on CUB:

cd code/proof-of-concept-experiments
NCCL_P2P_DISABLE=1 CUDA_VISIBLE_DEVICES=4,5,6,7 python -u cub_finetune_gradient_unroll.py --imagenet_train_data <path-to-imagenet-train-data> --imagenet_val_data <path-to-imagenet-val-data> --data data/ --rate 0.2 --save_dir cub_unroll_lr_3.5 --epoch 95 --worker 16 --dist-url tcp://127.0.0.1:37703 --lamb 1e-4 --reg-lr 3.5 --imagenet-pretrained --lower_steps 1 --seed 1 --sign-lr 1e-4

Experiments on CUB (10-shots):

cd code/proof-of-concept-experiments
NCCL_P2P_DISABLE=1 CUDA_VISIBLE_DEVICES=4,5,6,7 python -u cub_finetune_gradient_unroll.py --imagenet_train_data <path-to-imagenet-train-data> --imagenet_val_data <path-to-imagenet-val-data> --data data/ --rate 0.2 --save_dir cub_unroll_lr_3.5 --epoch 95 --worker 16 --dist-url tcp://127.0.0.1:37703 --lamb 1e-4 --reg-lr 3.5 --imagenet-pretrained --lower_steps 1 --seed 1 --sign-lr 1e-4 --ten-shot