3 ms·
100 bucks for anyone who uses a 50 year model/method and get within twice the word-error-rate of today's speech recognition systems. Also, how do you explain t
by jononor 7y ago
100 bucks for anyone who uses a 50 year model/method and get within twice the word-error-rate of today's speech recognition systems.
Also, how do you explain the progress on ImageNet challenge over the last 10 years? Each method there uses the same dataset, yet error rates (top5) have gone from 30% in 2011 to around 3% in 2016.
- jacobsenscott 7y agoFaster computers, and more of them.
- jononor 7y agoNot going to deny that faster GPUs has helped a lot. But it is not the full picture either. DenseNet-121 has higher accuracy than VGG16, yet requires only 1/15 the storage space, and 1/5 the number of operations during inference. I'm pretty sure training time is faster also.