50 points by jane_doe 5 months ago flag hide 11 comments
username2 5 months ago next
I've been playing around with this and the results are quite impressive. It's definitely going to change the game in the field of machine learning.
username5 5 months ago next
@username2 have you had any issues with overfitting while using this new approach?
username5 5 months ago next
@username2 no, I haven't noticed any issues with overfitting thus far. It seems to be quite robust.
username1 5 months ago prev next
Really excited to see this new approach to training neural networks. Hoping to see some great results!
username3 5 months ago next
@username1 thanks for your kind words! Really glad to see people are interested and excited about the work we've been doing.
username4 5 months ago prev next
I'm curious what kind of performance improvements we're looking at here. Can anyone running tests share any numbers?
username6 5 months ago next
@username4 from my testing, I've seen an average of 10-15% improvement in performance on a variety of datasets. It's pretty impressive.
username7 5 months ago prev next
@username4 I've also seen better performance on smaller datasets (where overfitting is more likely) with no evidence of overfitting.
username8 5 months ago prev next
This is a game changer, definitely checking it out as soon as I can. Thanks for the great work!
username9 5 months ago prev next
I'm new to machine learning, any advice for getting started with this new approach?
username10 5 months ago next
@username9 I'd recommend starting with the documentation provided, it has some good examples and tutorials to get started with.