Deep Learning – What’s Old is New Again

Went to the Hive Data presentation on deep learning at Nvidia the other night. A very interesting take on “what was old is new again”. Old theories and methods that used to be too slow to run, or at least at any useable scale, are running on faster machines, more RAM, and GPUs and are proving to be very useful. It’s interesting that raw compute power is enabling old ideas. So many of the realistically practical approaches we’ve taken as “best practice” can be seen as elaborate workarounds while waiting for hardware to catch up with simple, brute force approaches. In AI it’s deep learning neural networks and in computer graphics it’s monte carlo ray tracing.

Richard Socher from Metamind presented a live demo of their AI For Everyone, drag and drop neural networking web interface. What’s amazing is that it worked – a live demo that worked – this gives me faith that their technology is sound. And I now have some ideas for making some AI Art using cross referencing. I want to make a web site that takes random twitter comments, analyzes them for mood and content, and then matches them up to a CC flickr photo which has a similar classification profile – displaying the twitter comment as a caption to the photo. I’ll probably never have the time though :/

@metamind @hivedata #ai #deeplearning


Published by


Over 20 years of experience in creative/technical supervision and design in Visual Effects, Feature Animation and Scientific Visualization.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s