I think Blockchain technology as distributed immutable ledger is a very nice data structure for a lot of real world use cases (supply chain, market place, contracts,..). Sure it's also a huge hype thanks to crypto currencies, icos, etc. but the technology fulfills a need. However, I see more advantages in using private quorums leveraging proof of authority instead of public networks of untrusted nodes. I understand that not everyone might be ok with this tradeoff.
Distributed immutable logs of data existed long before the Blockchain (as someone pointed out, Git is a good example - just sign your commits). The innovation of the Blockchain was specifically the combination of that with the POW, which enabled coordination between untrusted nodes without falling prey to Sybil attacks.
A private blockchain is just giving another name to a preexisting model. It's like saying "I use a pedelec, but I see more advantages in removing the motor and battery".
Honestly, if you don't like the way certain Meetups are organised, the quality of the talks, etc. you can always help to make them better. Just prepare a talk such that most of the attendees can learn something from it. I am sure most Meetups will be happy to have you!
Here in Vienna we have a couple of hands-on Meetups that require you to bring your laptop and actually do something. Less people attend - but those that come are usually very motivated. This is a good place/way to learn something new.
I used the ami with a p2.large for a deep learning project at university. In my opinion it was quite pricey and reading from disk storage was disappointingly slow (we had a 60gb dataset). What i found really great was using the same machine first as a t2.micro instance for setup and testing s3 connection, a m4.large for some preprocessing and finally the p2 for deep learning, all on the same AMI and SSD.
Check out this blog post about deep learning for computer vision applications [1]; it describes very well the historical development and the current advances. In short, not too much changed from LeCuns ConvNets in the 80s.
Disclaimer, I wrote this article.
This should be possible and fairly easy to do, I guess - it's just forward and backward pass with your favorite model through the net and collect the loss for all octaves. You will have to port a few image utility functions (like roll, zoom, etc.) if not available. I ported the basic Deep Dream example to JavaScript [1] and it was not that difficult.
When looking through the code I saw that BP for LRN is not implemented yet, so you cannot pick a model using LRN (or you have to implement it).
I ported Caffe models to JavaScript in my free time http://chaosmail.github.io/projects/. however, since I know about Keras.js and it's GPU support, it doesn't seem that fancy anymore.
