2018-03-10 11:25 AM
Hi.
I watched CES 2018 Demo on youtube
I'm very interested in STM32CubeMX.AI.
How much ROM/RAM size are needed in library for neulral network?
How much CPU clock are needed for Inference in real time processing?Is it possible to use for MCU which don't have FPU?Can I try it?
Thanks.
2018-03-13 02:52 AM
Hello
tenderfoots366
,Many thanks for your interest on our STM32CubeMx.AI (Here you find the '
'presentation). The STM32CubeMX.AI is a client-server tool fully developed in ST. It makes automatic and fast conversion of a pre-trained neural network into a memory efficient and computationally fast optimized library.The library then can be executed on any STM32 MCU platform. The tool is planned for release later this year as a part of the STM32 software ecosystem. All details about it will be provided at that time.We thank you once more for your interest!
Khouloud.
2018-03-14 05:38 AM
Hello
tenderfoots366
,I have more details which could be useful for you.
Khouloud.
2018-03-14 07:31 AM
Hello Khouloud,
Thank for to answer my question.
I'm looking forward to the day when it will be open.
Best regard
Thanks.2018-03-14 08:07 AM
Thank you for sharing you work details related to AI.
I am really excited that ST is investing some effort in this area, and is going to help us developing AI-based apps.
If it is not a problem could you please share what sort of problems you are trying to solve with AI/STM32.
I watched the video where the STM32-based AI solution was used to differentiate the noise generated in house, outside, and to act accordingly, etc.
The computing power of the STM32 is limited that's why I am interesting in the problem space - if you could give some examples where the STM32/AI will be applicable. I do understand that most complex problems will be solved using external powerful servers, and the STM32 will be used as the front-end of the solution. But I am more interested in stand-alone apps.
Thanks in advance for any insights related to this topic.
2018-03-15 04:00 AM
Hi
tenderfoots366
,You're always welcome
We really appreciate your interest!
Khouloud.
2018-03-15 04:02 AM
Hi
Golab.Piotr
,We will share all needed information ASAP.
Just stay tuned
Khouloud.
2018-03-15 04:05 AM
I'll be studying machine learning while waiting;)
Thanks a lot.
2018-03-15 04:08 AM
garsi.khouloud
wroteThe STM32CubeMX.AI is a client-server tool fully developed in ST. It makes automatic and fast conversion of a pre-trained neural network into a memory efficient and computationally fast optimized library.The library then can be executed on any STM32 MCU platform.
I think the point of the '
pre-trained
neural network' is that all the 'heavy' training work & generating the Neural Network is doneon the server - not on the STM32 (the client)?So the STM32 doesn't do any learning itself.
2018-03-15 04:13 AM
I also suspect it that STM32 won't be involved because of the complexity of the problems where the AI is going to be used.
Trivial problems have decent solutions w/o AI;)