Neural Architecture Transfer

Neural architecture search (NAS) has emerged as a promising avenue for automatically designing task-specific neural networks. Existing NAS approaches require one complete search for each deployment specification of hardware or objective... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT BENCHMARK
Neural Architecture Search CIFAR-10 NAT-M1 Search Time (GPU days) 1.0 # 9
Parameters 4.3M # 15
FLOPS 232M # 11
Image Classification CIFAR-10 NAT-M2 Percentage correct 97.9 # 39
PARAMS 4.6M # 139
Image Classification CIFAR-10 NAT-M3 Percentage correct 98.2 # 31
PARAMS 6.2M # 140
Image Classification CIFAR-10 NAT-M1 Percentage correct 97.4 # 52
PARAMS 4.3M # 138
Image Classification CIFAR-10 NAT-M4 Percentage correct 98.4 # 24
PARAMS 6.9M # 141
Neural Architecture Search CIFAR-10 NAT-M4 Search Time (GPU days) 1.0 # 9
Parameters 6.9M # 18
FLOPS 468M # 16
Neural Architecture Search CIFAR-10 NAT-M3 Search Time (GPU days) 1.0 # 9
Parameters 6.2M # 17
FLOPS 392M # 15
Neural Architecture Search CIFAR-10 NAT-M2 Search Time (GPU days) 1.0 # 9
Parameters 4.6M # 16
FLOPS 291M # 12
Image Classification CIFAR-100 NAT-M4 Percentage correct 88.3 # 25
PARAMS 9.0M # 124
Neural Architecture Search CIFAR-100 NAT-M1 FLOPS 261M # 3
Percentage Error 14.0 # 5
PARAMS 3.8M # 3
Neural Architecture Search CIFAR-100 NAT-M2 FLOPS 398M # 4
Percentage Error 12.5 # 3
PARAMS 6.4M # 4
Neural Architecture Search CIFAR-100 NAT-M3 FLOPS 492M # 5
Percentage Error 12.3 # 2
PARAMS 7.8M # 5
Neural Architecture Search CIFAR-100 NAT-M4 FLOPS 796M # 6
Percentage Error 11.7 # 1
PARAMS 9.0M # 6
Image Classification CIFAR-100 NAT-M1 Percentage correct 86.0 # 37
PARAMS 3.8M # 120
Image Classification CIFAR-100 NAT-M2 Percentage correct 87.5 # 29
PARAMS 6.4M # 122
Image Classification CIFAR-100 NAT-M3 Percentage correct 87.7 # 28
PARAMS 7.8M # 123
Neural Architecture Search CIFAR-10 Image Classification NAT-M3 Percentage error 1.8 # 2
Params 6.2M # 11
FLOPS 392M # 14
Neural Architecture Search CIFAR-10 Image Classification NAT-M2 Percentage error 2.1 # 7
Params 4.6M # 8
FLOPS 291M # 13
Neural Architecture Search CIFAR-10 Image Classification NAT-M1 Percentage error 2.6 # 13
Params 4.3M # 7
FLOPS 232M # 12
Neural Architecture Search CIFAR-10 Image Classification NAT-M4 Percentage error 1.6 # 1
Params 6.9M # 12
FLOPS 468M # 15
Image Classification CINIC-10 NAT-M2 Accuracy 94.1 # 2