Video

Transcript

at Oba I our robotics team has been

working to bridge the reality gap that

separates simulated robotics from

experiments on hardware to this end we

have developed a novel technique for

transfer learning which allows us to

train a detector entirely in simulation

to spot objects and then to generalize

its knowledge to the physical world

we’re excited to demonstrate that this

technique can be applied to difficult

real-world problems such as spam

detection to demonstrate the

capabilities of our detectors we show

that they can be used for firm grasping

and cluttered environments they have

never seen before to our knowledge this

is the first successful transfer of a

deep neural network trained entirely in

simulation for the purpose of citizens

fan removal in the future we hope to

expand our capabilities to tackle

phishing attack we also plan to

experiment the generated adversarial

spam in order to improve the robustness

of our training data if you’d like to

find out more about our work visit

blogged up open a accom

[Music]