Pozorvlak on Nostr: TIL recent ARM "microcontrollers" have special instructions for vector arithmetic ...

TIL recent ARM "microcontrollers" have special instructions for vector arithmetic with dual numbers: https://developer.arm.com/documentation/101273/0001/The-Cortex-M55-Instruction-Set--Reference-Material/Armv8-1-M-vector-arithmetic-instructions/VQDMLADH--VQRDMLADH I know (thanks to nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqc9m22hkc5h6zgrwkz48crhcpw6vch2rf6j97746ugl3neys86jeqcr59k6 (nprofile…59k6)!) that dual numbers are useful for automatic differentiation and hence training neural networks by gradient descent, but who's doing NN *training* on embedded cores?