- Fire modules consisting of a 'squeeze' layer with 1*1 filters feeding an 'expand' layer with 1*1 and 3*3 filters(通過餵食一個包含1*1和3*3濾波器的'擴展'層,激勵包含一個'擠壓'層的模塊)
- AlexNet level accuracy on ImageNet with 50x fewer parameters(具有AlexNet水平的精度,卻少了50倍的參數量)
- Can compress to 510x smaller than AlexNet(0.5Mb)(可以比AlexNet壓縮510倍)