The paper "ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware" is a gradient based method.

D-X-Y / Awesome-AutoDL

Automated Deep Learning: Neural Architecture Search Is Not the End (a curated list of AutoDL resources and an in-depth analysis)

MIT License

2.23k stars 314 forks source link

The paper "ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware" is a gradient based method. #2

Closed zhengxiawu closed 5 years ago

D-X-Y commented 5 years ago

Thanks

Lyken17 commented 5 years ago

umm. I am the author of ProxylessNAS. Actually, we have experimented on both RL and Gradients.

D-X-Y commented 5 years ago

@Lyken17 Yes, I will correct it.

Lyken17 commented 5 years ago

By naively considering latency as (CE_loss + lambda * latency), the Gradients-based models are slightly worse than RL-based models. So our main results are from RL-based approaches.