Hi Arthur,

Amazing explanation. I was wondering, in the case of ResNets, if there are skip connections for backward gradient flow, there is a chance that a layer will be updated twice, one through skip connections and other through the usual backward path, right? Correct me if I am wrong.