Multi-Objective Neural Architecture Search
3M Seminar
Vishnu Boddeti
October 23, 2020
VishnuBoddeti
Not scalable to the increasing demand for AI solutions.
Too expensive to do for each desired operating point
Our solution: Population Based Multi-Objective Optimization
Our Approach: fine-tune supernet
Method | Year() | #MAdds() | #Params() | Test AUROC() |
---|---|---|---|---|
NIH | 2017 | - | - | 73.8% |
CheXNet | 2017 | 2.8B | 7.0M | 84.4% |
Google AutoML | 2019 | - | - | 79.7% |
LEAF | 2019 | - | - | 84.3% |
NSGANet-A3 | 2019 | 0.6B | 5.0M | 84.7% |
NSGANet-X | 2019 | 0.4B | 2.2M | 84.60% |
MUXNet-m | 2020 | 0.2B | 2.1M | 84.10% |
Network | #MAdds() | $AP_{bbox}$ | $AP_{mask}$ |
---|---|---|---|
MobileNetV3 | 219M | 44.02% | 43.60% |
FBNetV2 | 238M | 44.83% | 43.86% |
NAT-M1 | 225M | 45.23% | 44.27% |
Network | #MAdds() | #Params() | mIoU() | Acc() |
---|---|---|---|---|
ResNet18+C1 | 1.8B | 11.7M | 33.82% | 76.05% |
MobileNetV2+C1 | 0.3B | 3.5M | 34.84% | 75.75% |
MUXNet-m+C1 | 0.2B | 3.4M | 32.42% | 75.00% |
ResNet18+PPM | 1.8B | 11.7M | 38.00% | 78.64% |
MobileNetV2+PPM | 0.3B | 3.5M | 35.76% | 77.77% |
MUXNet-m+PPM | 0.2B | 3.5M | 35.80% | 76.33% |
Network | #MAdds() | Cityscapes | PASCAL VOC | COCO-Stuff |
---|---|---|---|---|
MobileNetV3 | 219M | 73.0% | 73.8% | 28.5% |
FBNetV2 | 238M | 72.6% | 73.6% | 28.5% |
NAT-M1 | 225M | 74.0% | 75.9% | 29.5% |
Network | #MAdds() | #Params() | $mAP$() |
---|---|---|---|
VGG16+SSD | 35B | 26.3M | 74.3% |
MobileNet+SSD | 1.6B | 9.5M | 67.6 |
MobileNetV2+SSDLite | 0.7B | 3.4 | 67.4% |
MUXNet-m+SSDLite | 0.5B | 3.2M | 68.6% |
MUXNet-l+SSD | 1.4B | 9.9M | 71.1% |
Network | #MAdds() | $mAP$() | $AP_{0.5}$() |
---|---|---|---|
MobileNetV3 | 219M | 31.8% | 49.8% |
FBNetV2 | 238M | 31.1% | 48.8% |
NAT-M1 | 225M | 32.2% | 50.4% |