Page 54 - GCN, June/july 2017

t
WishLis
Tech we hope to see in the public sector
Machine learning as a service
It might look like a city skyline, but this tensor processing unit is Google’s latest chip optimized for machine-learning tasks. Unveiled in May, the second- generation TPU can deliver up to 180 teraflops of performance and can also
be used for training machine-language models. It will soon be available as a service on the Google Cloud Platform.
A truly quick study
A new robot-teaching system developed by the nonprofit OpenAI research company promises “one-shot imitation learning.” A human demonstrates a task, such as stacking blocks,
in a virtual reality environment. A commercially available robot can then successfully stack almost any array of blocks after a single lesson, thanks to two neural networks — one to “observe” the demonstration and another to control the movements.
Smarter dummies
The latest version of the Army Research Laboratory’s Warrior Injury Assessment Manikin relies on advanced sensors and carefully engineered bio-fidelity to measure the potential effects of explosions with unprecedented accuracy. Currently used to assess survivability in the design of new military vehicles, the dummies have a wide range of safety and simulation applications.
What new technologies do you think GCN readers should learn more about? Tell us on Twitter: @GCNtech #GCNwishlist.
Image credits: Google, U.S. Army, Fetch Robotics