Sinisterly
Intro to deep learning - Printable Version

+- Sinisterly (https://sinister.ly)
+-- Forum: Coding (https://sinister.ly/Forum-Coding)
+--- Forum: Python (https://sinister.ly/Forum-Python)
+--- Thread: Intro to deep learning (/Thread-Intro-to-deep-learning)



Intro to deep learning - Inori - 02-06-2017

AI, machine learning, and deep learning/neural networks are topics that I've been interested in recently, but I've found a lack in tutorials and introductions. I just found this repo and I plan on poking around and seeing what I can make with it.

https://github.com/yala/introdeeplearning


RE: Intro to deep learning - silur - 02-06-2017

Deep learning is a hype train I cannot understand. Not only there is proof that not all computationally feasible problems can be solved with deep learning (however can be solved by a human), but it has drawbacks, overhead and science institutions highly tend to avoid neural networks with AI.
There are much more sophisticated methods even if they are more specific to a certain task, so I just can't understand why people hype DNN so hard, they even want to calculate logarithms with neural networks T_T


RE: Intro to deep learning - Inori - 02-06-2017

(02-06-2017, 04:40 PM)silur Wrote: Deep learning is a hype train I cannot understand. Not only there is proof that not all computationally feasible problems can be solved with deep learning (however can be solved by a human), but it has drawbacks, overhead and science institutions highly tend to avoid neural networks with AI.
There are much more sophisticated methods even if they are more specific to a certain task, so I just can't understand why people hype DNN so hard, they even want to calculate logarithms with neural networks T_T

Yes, deep learning and neural networks aren't an exact science (yet), but they're a relatively new frontier in computer science. Given time, they'll likely improve in accuracy and become more accessible to the average programmer. Speaking of new horizons, think about how awesome neural networks could be on a quantum processor Tongue


RE: Intro to deep learning - Hoss - 02-09-2017

(02-06-2017, 04:40 PM)silur Wrote: Deep learning is a hype train I cannot understand. Not only there is proof that not all computationally feasible problems can be solved with deep learning (however can be solved by a human), but it has drawbacks, overhead and science institutions highly tend to avoid neural networks with AI.
There are much more sophisticated methods even if they are more specific to a certain task, so I just can't understand why people hype DNN so hard, they even want to calculate logarithms with neural networks T_T

I don't know... I think it's more than a hype. I've personally used neural-networks for packet scanning and HR Q/A automation at work and once we had the NN trained we "compiled" them to x86 assembly and they were extremely efficient with a high accuracy rate. I'll see if I can dig up some of the code for one of those projects and release it here.

And yes of course, NNs are not ALWAYS going to be the solution you need, but I think for a lot of scenarios a NNs dynamic nature really comes in handy.


RE: Intro to deep learning - Satan - 02-09-2017

(02-09-2017, 06:40 AM)Hoss Wrote:
(02-06-2017, 04:40 PM)silur Wrote: Deep learning is a hype train I cannot understand. Not only there is proof that not all computationally feasible problems can be solved with deep learning (however can be solved by a human), but it has drawbacks, overhead and science institutions highly tend to avoid neural networks with AI.
There are much more sophisticated methods even if they are more specific to a certain task, so I just can't understand why people hype DNN so hard, they even want to calculate logarithms with neural networks T_T

I don't know... I think it's more than a hype. I've personally used neural-networks for packet scanning and HR Q/A automation at work and once we had the NN trained we "compiled" them to x86 assembly and they were extremely efficient with a high accuracy rate. I'll see if I can dig up some of the code for one of those projects and release it here.

And yes of course, NNs are not ALWAYS going to be the solution you need, but I think for a lot of scenarios a NNs dynamic nature really comes in handy.

extremely limited situations where they would be handy, generally.

I just think they're a quirky thing to do, as far as NN, as can be seen with the most practical application I've seen of it being this:



RE: Intro to deep learning - Blink - 02-09-2017

Hmm... I used to be very interested in machine learning (still am, just not to the same extent)

A good intro to machine learning is what I found here: https://medium.com/@ageitgey/machine-learning-is-fun-80ea3ec3c471#.u5ry3mij2