This repository accompanies the paper Towards Tailored Models on Private AIoT Devices: Federated Direct Neural Architecture Search and contains the basic code for replicating the training and computations in it. We will continue to add more details and code to this repository as we go along
nohup python federated_main.py --gpu 2 --search --warmup --resume --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 128 --test_batch_size 128 --init_lr 0.025 --arch_algo grad --arch_lr 0.025 > 1.out 2>&1 &nohup python federated_main.py --gpu 0 --search --resume --last_round 175 --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 128 --test_batch_size 128 --init_lr 0.025 --arch_algo grad --arch_lr 0.025 --target_hardware cpu --object_to_search cpu --grad_binary_mode two --grad_reg_loss_type add#linear --grad_reg_loss_lambda 0.05 > cls_cpu.out 2>&1 &
nohup python federated_main.py --gpu 2 --search --resume --last_round 175 --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 128 --test_batch_size 128 --init_lr 0.025 --arch_algo grad --arch_lr 0.025 --target_hardware gpu8 --object_to_search gpu8 --grad_binary_mode two --grad_reg_loss_type add#linear --grad_reg_loss_lambda 0.05 > cls_gpu.out 2>&1 &nohup python convert_normal_architecture.py --gpu 0 --path output/exp_name/ --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ > cvt.out 2>&1 &nohup python retrain.py --path output/fednas-grad12,3,4,3,4,3/learned_net --gpu 3 --resume --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 256 --test_batch_size 256 --init_lr 0.05 > 2.out 2>&1 &
nohup python retrain.py --resume --path output/fednas-grad1cpu2,3,4,3,4,3/learned_net --object_to_retrain cpu --gpu 0 --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 128 --test_batch_size 128 --init_lr 0.025 --last_round 300 > 160to250round_cpu_retrain.out 2>&1 &
nohup python retrain.py --resume --path output/fednas-grad1gpu2,3,4,3,4,3/learned_net --object_to_retrain gpu8 --gpu 0 --dataset_location /home/grouplcheng/data/zch/dataset/cifar100/ --train_batch_size 128 --test_batch_size 128 --init_lr 0.025 --last_round 300 > 160to250round_gpu_retrain.out 2>&1 &
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent