Accelerating Understanding: Deep Learning, Intelligent Applications, and GPUs

The Institute for Scientific Computing Research (ISCR) sponsored this talk entitled "Deep Learning" on April 16, 2015, at the Lawrence Livermore National Laboratory. The talk was presented by Yann LeCun, director of AI research at Facebook and professor of data science, computer science, neural science and electrical engineering at NYU.
Learn more about ISCR: computation.llnl.gov/about/org...

[@ncroc]

Interesting thing is 1-2 years later neural networks will be abandoned again. Most people will go back to using probabilistic models again. Why ? Because of unlabeled data. Wait and see.

[@CM-us8tu]

ok, I will wait and see. You said 2 years right? OK, I got time. In 2 years, I will come back here and comment on the status of neural networks

[@ncroc]

Hopefully it won't take that long but you know it takes a long time for people to quit bad habits :)

[@CM-us8tu]

I agree, but it seems that artificial neural network according to experts, it gets better with more data and as you already know, through cloud computing data can be manipulated easily. It is said that only artificial neural network can cope with exponential data humans are creating and the algorithms gets better with more data. It might seem that it has potential.

[@ncroc]

There are tons of problems with neural netoworks but the main problem is that they need tons of LABELED data. Some might call them data hungry but lets face it, they are pretty stupid. If they need so much supervision, it is obvious that they are not even artificially intelligent. You will see that probabilistic models (which can used for supervised, unsupervised and semisupervied learning ) that can make efficient use of unlabeled data will dominate the field in the near furute (1-2 years). Dont believe the hype.

[@MrGenbu]

HMMMM !!!

[@shengyuan1840]

Deep Insight