XPost: comp.misc, alt.microcontrollers, comp.arch.embedded   
   XPost: alt.microcontrollers.8bit   
   From: d@ray   
      
   Bouyed by the surprisingly good performance of neural networks with   
   quantization aware training on the CH32V003, I wondered how far this can be   
   pushed. How much can we compress a neural network while still achieving   
   good test accuracy on the MNIST dataset? When it comes to absolutely   
   low-end microcontrollers, there is hardly a more compelling target than the   
   Padauk 8-bit microcontrollers. These are microcontrollers optimized for the   
   simplest and lowest cost applications there are. The smallest device of the   
   portfolio, the PMS150C, sports 1024 13-bit word one-time-programmable   
   memory and 64 bytes of ram, more than an order of magnitude smaller than   
   the CH32V003. In addition, it has a proprieteray accumulator based 8-bit   
   architecture, as opposed to a much more powerful RISC-V instruction set.   
      
   Is it possible to implement an MNIST inference engine, which can classify   
   handwritten numbers, also on a PMS150C?   
      
   …   
      
   …   
      
      
      
      
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)