AdamW and Super-convergence is now the fastest way to train neural nets

Re:AdamW and Super-convergence is now the fastest way to tra... by sandeepeecs on Wednesday July 04, 2018 @02:01AM (#17455513)

Nice to see great advancements in the training speed for deep neural network algorithms. We are also very interested in improving the speed of training. We at alpes.ai have developed a non-recursive neural network algorithm which has a very fast training time. We got very good results for standard open datasets, training time in the range of seconds and accuracy on par with standard results on normal laptops without any special GPU`s or hardware.

http://alpes.ai/

These are the results of some of the datasets

DataSet Training Time Accuracy

Extended Yale DB 40 sec 94.00%

Human Activity Detection Dataset 3 sec 86%

MNIST 90 sec 97%

Google Speech Dataset 60 sec 92%

Liver Dataset 2 sec 89%

[ Parent | on HN ]

Re:AdamW and Super-convergence is now the fastest way to tra... by cnewey on Wednesday July 04, 2018 @08:47AM (#17456947)

> We believe, the SNN is the correct representation of the how the mind works. In fact, we have mathematical proof that our representation is the most efficient representation.

No details on the website, though. This is either very poor marketing, or (much more likely) snake oil.

[ Parent | on HN ]

Re:AdamW and Super-convergence is now the fastest way to tra... by osaariki on Wednesday July 04, 2018 @02:35AM (#17455620)

The page you linked is very light on details. Could you point to a paper?

[ Parent | on HN ]

Re:AdamW and Super-convergence is now the fastest way to tra... by sandeepeecs on Wednesday July 04, 2018 @09:02AM (#17457016)

hey paper is on its way probably we will be publishing it in couple of weeks from now.

[ Parent | on HN ]

Re:AdamW and Super-convergence is now the fastest way to tra... by yorwba on Wednesday July 04, 2018 @11:01AM (#17457617)

Make sure to post it to HN so we can criticize your work into the ground ;)

[ Parent | on HN ]