From Wikipedia.
"In computing, especially digital signal processing, the multiply–accumulate or multiply-add operation is a common step that computes the product of two numbers and adds that product to an accumulator"
Does this whole debate, come down to you saying that our little boy AKIDA, can't do simple maths?..
..
..
I'm saying it doesn't need to - as Ella reminds us "Tain't what you do - it's the way that you do it!"
This is an XL (originally Lotus) 4-bit binary* digital MAC simulator that I invented many years ago:
*The secret sauce is in the hidden formulae and the use of decimal numbers (black partial sums).
Basically MAC matrix maths multiplies rows and columns. Until very recently the binary numbers would commonly have been 32 bits or more. So at least 64 times (8*8) the above just to multiply a pair of numbers. Hence the brilliance of PvdM's initial 1-bit concept which still has many use cases.
Note: This drawing is not an example of Akida's NPU - see the patent drawing above.
However, when the initial 1-bit FPGA was field trialled by early adopters, a need for greater accuracy was highlighted for some use cases. Hence Akida 1 was developed with 4-bit weights and activations. with this change, I asked PvdM if that would mean that Akida 1 would use MACs. His reply was "No".
So perhaps, PvdM should be corrected on this point.
www.linkedin.com
