xiaomi-automl / SCARLET-NAS

Bridging the gap Between Stability and Scalability in Neural Architecture Search
142 stars 14 forks source link

Retrain the searched model #3

Open xingyueye opened 5 years ago

xingyueye commented 5 years ago

Great ideas. I want to know whether retraining is required after the search stage and remove of the ID operation. Does it has an influence on the characteristic expression ability of the network before and after dropping the Conv in LET?