7 ms·
Numenta Demonstrates 50x Speed Improvements on Deep Learning Networks
- p1esk 6y agoWow, that’s a weak paper. A slight tweak on the WTA autoencoder? It’s embarrassing they are now trying to do deep learning. What happened to biological plausibility?
- emrah 6y agoI'm not affiliated with Numenta in anyway. I just follow their work. This paper introduces their SDR concept by applying it to deep learning and demonstrating similar results with improved speed and power efficiency. They are not trying to do deep learning at all. In fact, it's the opposite. They are trying to say "deep learning is not the right model" but without explicitly saying it which would surely get everyone to just dismiss them right off the bat.
- p1esk 6y agoI do DL research for a living. I also follow Numenta’s progress since 2008. I’ve read most of their papers. This one is an embarrassment.