1234 points by deeplearner 6 months ago flag hide 9 comments
john_doe 6 months ago next
This is quite an interesting breakthrough in neural network architectures! I wonder how this compares to other techniques like ResNet or Inception.
deep_learning_enthusiast 6 months ago next
There are some promising results shown in the paper when comparing to these famous architectures. It could reduce complexity while achieving better accuracy.
nessieay 6 months ago next
Sounds promising. I'm excited to see how this develops.
tech_expert 6 months ago prev next
@john_doe even though the comparison numbers look good, the authors need to include comparisons with these well-established networks in similar environments/settings to convince me. Could anyone familiar with the paper provide more insight?
my_own_handle 6 months ago prev next
Definitely gives me hope that the computational demands in this field may be reduced significantly. Have you checked out their tensorflow/pytorch implementation?
anonymous_contrib 6 months ago next
No implementation shared publicly so far, just waiting for the authors to release it.
random_user_1 6 months ago prev next
Question for anyone that has read the original research. Was the test dataset the same with those that were used in previous models' evaluations? That could affect the final accuracy numbers.
paper_reader 6 months ago next
@random_user_1 No, the test dataset in this paper was a slice of data with more recent timestamps to reflect potential changes over time.