The DeepMind and the Human Mind
January 7, 2015
As advanced as computers are, they still fail to replicate the human brain’s complexity. Technology Review’s article “Google’s Secretive DeepMind Startup Unveils A ‘Neural Turing Machine’” tells the story about how Google bought the startup DeepMind that is working on a prototype computer to mimic a human’s short-term working memory. The prototype is a neutral network that learns from its “memories” and then later retrieves them to perform logical tasks.
DeepMind named its project the Neural Turing Machine and it takes the basic premise of a neural network, interconnected “neurons” that respond to external input, and it acts more like a computer by storing information and retrieving it. DeepMind is testing the Neural Turing Machine by feeding it specific tasks and then giving it more complex ones.
The prototype is doing much better than a typical neural network:
“They compare the performance of their Neural Turing Machine with a conventional neural network. The difference is significant. The conventional neural network learns to copy sequences up to length 20 almost perfectly. But when it comes to sequences that are longer than the training data, errors immediately become significant. And its copy of the longest sequence of length 120 is almost unrecognizable compared to the original.”
Neural networks are closer to replicating the human brain, but they are still not close to understanding how humans work through arguments and handle high-level tasks.
Whitney Grace, January 07, 2014
Sponsored by ArnoldIT.com, developer of Augmentext
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
