Microsoft Open Sources Distributed Machine Learning Toolkit | Hacker News Reader

Microsoft Open Sources Distributed Machine Learning Toolkit

dmtk.io

129 pointskostandin_k15 comments

https://news.ycombinator.com/item?id=10542004

After quickly browsing the docs how do you interact with this? It's written in C++ and needs ZeroMQ. That's about all I managed to get out of it so far, no code examples and very light docs.

maybe they open-sourced that project in a rush, following Google's move?

That occurred to me too, but the dmtk.io domain was registered a couple of months ago on 2015-09-08.

I had the same feeling, the website doesn't look ... done. However, I think it is nice to see that MS is launching its own initiative. The more, the marrier. I'm especially interested to see how their LDA variant performs.

Still, I hope they continue to make the documentation a bit more accessible. In example, the LightLDA documentation (http://www.dmtk.io/lightlda.html) gives no indication of the output I would get from this. Do I get all the matrices (word to topic, topic to doc) or just a list of top topics, or...?

Perhaps they planned to do it already, but they had to move their timetable up in reaction to Google's own efforts. If they felt giving google too much time might create a significant first mover advantage it could certainly make sense.

I am now left to wonder if IBM Watson will soon follow suit. IBM is really a very well constructed consultancy company.. I wonder if they plan to keep their IP for Machine Learning proprietary or if they see an opportunity to get more people using their technology (it has the better marketing name right now) and sell their services to a larger audience. Time will tell, but it is a very interesting time indeed.

no way, IBM is seeking to monetize it and they have good marketing lead already, everybody has heard of the chess story, but not many even heard about machine learning.

A lot of that going around lately...

In 100 years we may look back on this last month as the time when 'the machines were born'

I've been off grid the last 3 weeks. Could you throw in some words I could google up?

//Thank you.

TensorFlow (by Google): open-source library for machine intelligence

https://news.ycombinator.com/item?id=10532957

http://tensorflow.org/

@tommoor is making reference to Google and Microsoft releasing open source versions of their deep learning toolkits. Interesting that Google did not release a distributed version but Microsoft did.

As this stuff is more accessible, more people play with it, and perhaps more interesting learning apps emerge from that additional use. Hence, birth of the machines.

Microsoft has been pushing really hard to make Azure a contender. And they're right to do so, AWS is in a somewhat shakey position. They're moving people over to VPCs but haven't integrated their VPCs with their cloud services in a robust way, forcing the burden onto customers.

Anecdotally, friends of mine in the know tell me that it's been a good year for GCE. I wonder if Azure is seeing an uptick as well.

Also Samsung's platform "Veles" ... although it turns out this has actually been released for a couple of months:

https://velesnet.ml/

Reply on news.ycombinator.com