This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Peter Wells Memorial Lecture - continue the discussion here!

Huge thanks to Dr Kia Nazarpour for his fantastic presentation on 'The Ups and Downs of Machine learning for Prosthetic Control.'


If you missed the webinar you'll be able to catch up On Demand very soon.


It was clearly a very interesting topic for lots of you as we received a huge number of questions! Unfortunately we were unable to get through all of them but Kia is hoping to follow up on these and will respond below so please keep an eye out here. :)
  • Kenneth asked - 


    In the future is it likely that electrical information directly  from the brain rather than the limb will be more useful?
  • William asked - 


    Are the ML algorithms running in a computer or an edge device or cloud currently? How is it comparable to coapt?
  • Richard asked -


    Are you likely to be using 5G mobile technology as part of your human-in-the loop-AI data. collection?
  • Martin asked - 


    If we implant a net made of electrodes at the surface of the brain, can we read brainwaves from there and send them to the hand?
  • Joe asked - 


    With a single limb – is it possible to learn commands from the non-affected limb to speed up the training of the AI prosthetic?
  • Jasmine asked -


    For the prosthesis in the second video, what data was extracted from the IMUs? And was this data used alongside the EMG signals in the machine learning models?
  • Sarah asked - 


    Considering the low uptake of upper limb prosthesis in congenital users, do you believe this will increase uptake in this group and therefore increase the potential market and make it more economically viable research?
  • Keith asked -


    Are we now at an early stage such as heart transplants of 60 years ago?
  • Saeed asked - 


    What do you think of MIT magnetic sensors?
  • Vincent asked - 


    Your examples used 'pseudo-hands'.  Are there better/simpler configurations to simplify the AI challenge?  Make it more reliable and acceptable to the user?