script for all experiment

Baseline

1. DVSC10

python main.py --model VGG_SNN --node-type LIFNode --dataset dvsc10 --step 10 --batch-size 120 --act-fun QGateGrad --device 0 --seed 42 --DVS-DA  --traindata-ratio 1.0 --smoothing 0.0

2. NCALTECH101

python main.py --model VGG_SNN --node-type LIFNode --dataset NCALTECH101 --step 10 --batch-size 120 --act-fun QGateGrad --device 0 --seed 42 --num-classes 101 --traindata-ratio 1.0 --smoothing 0.0

3. omniglot

python main.py --model SCNN --node-type LIFNode --dataset nomni --step 12 --batch-size 16 --num-classes 1623 --act-fun BackEIGateGrad --epochs 200 --device 3 --log-interval 500 --smoothing 0.0 --lr 1.2 --sched step --decay-epochs 50 --opt adamw --warmup-epochs 5 --seed 52

4. EsimageNet

CUDA_VISIBLE_DEVICES=4,5,6,7 python -m torch.distributed.launch --nproc_per_node=4 --master_port=12355 main.py --model resnet18 --node-type LIFNode --dataset esimnet --step 8 --batch-size 32 --act-fun QGateGrad --seed 42 --DVS-DA --traindata-ratio 1.0 --num-classes 1000 --lr 5e-2 --warmup-epochs 1 --smoothing 0.0 --sched step --decay-epochs 6

Transfer with two loss

1. DVSC10

python main_transfer.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset cifar10 --target-dataset dvsc10 --step 10 --batch-size 120 --act-fun QGateGrad --device 1 --seed 1024 --traindata-ratio 1.0 --smoothing 0.0 --domain-loss --semantic-loss --DVS-DA

2. NCALTECH101

python main_transfer.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset CALTECH101 --target-dataset NCALTECH101 --step 10 --batch-size 120 --act-fun QGateGrad --device 2 --seed 47 --num-classes 101 --traindata-ratio 1.0 --smoothing 0.0 --domain-loss --semantic-loss --semantic-loss-coefficient 0.001

3. omniglot

python main_transfer.py --model Transfer_SCNN --node-type LIFNode --source-dataset omni --target-dataset nomni --step 12 --batch-size 16 --num-classes 1623 --act-fun BackEIGateGrad --epochs 200 --device 3 --log-interval 500 --smoothing 0.0 --lr 1.2 --sched step --decay-epochs 50 --opt adamw --warmup-epochs 5 --seed 52 --domain-loss --semantic-loss

Transfer with two loss and tet loss

1. DVSC10

python main_transfer.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset cifar10 --target-dataset dvsc10 --step 10 --batch-size 120 --act-fun QGateGrad --device 1 --seed 42 --traindata-ratio 1.0 --smoothing 0.0 --domain-loss --semantic-loss --DVS-DA --TET-loss-first --TET-loss-second

2. NCALTECH101

python main_transfer.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset CALTECH101 --target-dataset NCALTECH101 --step 10 --batch-size 120 --act-fun QGateGrad --device 5 --seed 42 --num-classes 101 --traindata-ratio 1.0 --domain-loss --semantic-loss --semantic-loss-coefficient 0.001 --TET-loss-first --TET-loss-second&

TET Results

Dataset	Model	Node-Type	Step	Act-Fun	Label Smoothing	TET Loss None	TET Loss First Only	TET Loss ALL
Dvsc10	VGG_SNN	LIFNode	10	QGateGrad	None	81.3%	82.4%	82.9%
NCALTECH101	VGG_SNN	LIFNode	10	QGateGrad	None	76.2%	78.4%	80.8%

Visualization Loss-landscape

HDF5_USE_FILE_LOCKING="FALSE" mpirun -n 4 -mca btl ^openib python main_visual_losslandscape.py --model VGG_SNN --node-type LIFNode --source-dataset cifar10 --target-dataset dvsc10 --step 10 --batch-size 1000 --eval --eval_checkpoint /home/hexiang/TransferLearning_For_DVS/Resultes_new_compare/Baseline/VGG_SNN-dvsc10-10-seed_42-bs_120-DA_True-ls_0.0-traindataratio_0.1-TET_first_False-TET_second_False/last.pth.tar --mpi --x=-1.0:1.0:51 --y=-1.0:1.0:51 --dir_type weights --xnorm filter --xignore biasbn --ynorm filter --yignore biasbn --plot --DVS-DA --smoothing 0.0 --output /home/hexiang/TransferLearning_For_DVS/Resultes_new_compare/ --traindata-ratio 0.1

python main_visual_losslandscape.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset CALTECH101 --target-dataset NCALTECH101 --step 10 --batch-size 500 --eval --eval_checkpoint /home/hexiang/TransferLearning_For_DVS/Results_new_compare/train_TCKA_test/Transfer_VGG_SNN-NCALTECH101-10-bs_120-seed_47-DA_False-ls_0.0-lr_0.005-SNR_0-domainLoss_True-semanticLoss_True-domain_loss_coefficient1.0-semantic_loss_coefficient0.001-traindataratio_0.1-TETfirst_True-TETsecond_True/last.pth.tar --mpi --x=-1.0:1.0:51 --y=-1.0:1.0:51 --dir_type weights --xnorm filter --xignore biasbn --ynorm filter --yignore biasbn --plot --smoothing 0.0 --output /home/hexiang/TransferLearning_For_DVS/Results_new_compare/ --traindata-ratio 0.1 --num-classes 101 --device 5&

python main_visual_losslandscape.py --model VGG_SNN --node-type LIFNode --source-dataset CALTECH101 --target-dataset NCALTECH101 --step 10 --batch-size 500 --eval --eval_checkpoint /home/hexiang/TransferLearning_For_DVS/Results_new_compare/Baseline/VGG_SNN-NCALTECH101-10-seed_47-bs_120-DA_False-ls_0.0-lr_0.005-traindataratio_0.1-TET_first_True-TET_second_True/last.pth.tar --mpi --x=-1.0:1.0:51 --y=-1.0:1.0:51 --dir_type weights --xnorm filter --xignore biasbn --ynorm filter --yignore biasbn --plot --smoothing 0.0 --output /home/hexiang/TransferLearning_For_DVS/Results_new_compare/ --traindata-ratio 0.1 --num-classes 101 --device 5&

python main_visual_losslandscape.py --model Transfer_VGG_SNN --node-type LIFNode --source-dataset cifar10 --target-dataset dvsc10 --step 10 --batch-size 1000 --eval --eval_checkpoint /home/hexiang/TransferLearning_For_DVS/Results_new_compare/train_TCKA_test/Transfer_VGG_SNN-dvsc10-10-bs_120-seed_47-DA_True-ls_0.0-lr_0.005-SNR_0-domainLoss_True-semanticLoss_True-domain_loss_coefficient1.0-semantic_loss_coefficient0.5-traindataratio_0.1-TETfirst_True-TETsecond_True/last.pth.tar --mpi --x=-1.0:1.0:51 --y=-1.0:1.0:51 --dir_type weights --xnorm filter --xignore biasbn --ynorm filter --yignore biasbn --plot --DVS-DA --smoothing 0.0 --output /home/hexiang/TransferLearning_For_DVS/Results_new_compare/ --traindata-ratio 0.1 --device 0&

python main_visual_losslandscape.py --model VGG_SNN --node-type LIFNode --source-dataset cifar10 --target-dataset dvsc10 --step 10 --batch-size 1000 --eval --eval_checkpoint /home/hexiang/TransferLearning_For_DVS/Results_new_compare/Baseline/VGG_SNN-dvsc10-10-seed_47-bs_120-DA_True-ls_0.0-lr_0.005-traindataratio_0.1-TET_first_True-TET_second_True/last.pth.tar --mpi --x=-1.0:1.0:51 --y=-1.0:1.0:51 --dir_type weights --xnorm filter --xignore biasbn --ynorm filter --yignore biasbn --plot --DVS-DA --smoothing 0.0 --output /home/hexiang/TransferLearning_For_DVS/Results_new_compare/ --traindata-ratio 0.1 --device 7&

Record Id

CAT	131, 152, 94
Airplane	110, 164( ->108), 133
Bird	101, 191,151
Dog	56, 80, 20

N-omniglot

python main.py --model SCNN --node-type LIFNode --dataset nomni --step 12 --batch-size 64 --num-classes 1623 --act-fun QGateGrad --epochs 200 --device 6 --log-interval 200 --smoothing 0.0 --seed 42 --TET-loss-first --TET-loss-second