3 things direct from the future

23rd Edition

Once per week I will deliver “3 things direct from the future”. A 2 minute read that will always give you:

  • one thing that can help,
  • one thing to be wary of, and
  • one thing to amaze.

If this sounds interesting to you then please subscribe.

Subscribe

1. One thing that helps

Zoom Sign-Language

The true glory of the video conference call should be in being available to all, right? We need to make sure everyone can get busted not paying attention or have their cat walk in front of the camera.  Whatever we think of it, the video conference call is now an essential part of our lives. If you can’t fully participate, this will affect your ability to perform your job or communicate with friends and family.

A key function of these tools is that when you speak, the software highlights you as the speaker.  Deaf people, or those who have trouble speaking, find such video conferencing apps tedious, if not completely unusable. Google is now providing the ability to recognise WHEN a person is using sign-language using a technology they call PoseNet (someone seriously needs to stop tech people naming these things).

PoseNet grabs each frame of the video and breaks it down to a few factors (similar to the video conferencing real-time key points from last week) including a person’s eyes, nose, shoulders, and hands. Then it compares those factors, in real-time,  to a video database of people performing sign-language with 91.5% match accuracy! If it detects someone signing, it plays an ultrasonic tone to make the video call software spotlight that person. Most importantly, this algorithm is lightweight and can be deployed by any video call software.

A complex solution executed simply!

2. One to be wary of

Camouflaged Computer Chips

We all know about hacking of software systems. Turns out, even our computer’s brain – its chip- is also a target for hackers. By identifying which “type” transistors are in a circuit, someone can reverse engineer the transistor itself. This causes major headaches for security and for protecting intellectual property.  To prevent this, a team at Purdue University have used black phosphorus to camouflage transistors to look like they are the same type.

When voltage is applied to transistors, they show up as either N or P type based on the method they use to carry a current. The Purdue team have proven that using an extremely thin material such as black phosphorus camouflages the transistors so they appear the same to the hacker. They have also introduced a security key into the transistors which is needed to determine the type of transistor.

How important is this disguise? It protects the computer chip from being duplicated or exploited by hackers. Given access to the right tools, someone can figure out exactly what (and how) a system is doing if they know the composition of the chip. There is still work to be done, with black phosphorus too volatile for commercial use, but this research is paving the way for improved computer chip security.

3. One to amaze

 

Don’t Push the RoboDog!

If you’ve ever tried teaching your dog new tricks, you know the power of reinforcement learning – giving it a treat every time it does the trick correctly. Now, this same method is being used to train robot dogs.

Instead of writing lines of codes for every possible scenario (which would be an insurmountable task) a team from Zhejiang University and University of Edinburgh deployed neural networks to teach Jueying the robot dog. It has eight “expert” algorithms that specialize in a particular skill such as trotting, balancing and uprighting itself when it falls down. These eight algorithms work together and when faced with a particular situation the “coach”, or overarching network, will decide which of these expert algorithms takes over.

The robot dog is initially trained in a virtual world and presented with various obstacles. If it fails, it gets a demerit. If it succeeds, it is given a digital merit as a reward, much like how you train a dog. By trial and error, the robot dog learns and adapts to situations it has not encountered before. This means developers do not have to train the robot to respond to every situation it may find itself in. This enables the dog to learn how to walk over unstable ground, and fall and get up again no matter what position it may find itself in.

I am not sure how smart the people pushing them over in the video want these robot dogs to get though.  If I was a recently “self-aware” robot the first person I would be coming for is the one who pushed me over a thousand times.  OK maybe this one is in the wrong category this week!

Have a great week.

Daniel J McKinnon

Connect on LinkedIn

Subscribe

Subscribe

Don’t Stop Here

More To Explore