My MATHS teacher used to say ,

“Learn and Understand MATRIX otherwise “you will play computer games, but you will not be able to design one. “

And the irony today is, “the GAME(reinforcement learning) itself is redefining the matrix operations.”

Yes, I am talking about the DeepMind’s published paper (October 2022) in NATURE that describes their AI System Alpha Tensor which works on improving efficiency of fundamental computations algorithms.

https://deepmind.google/discover/blog/discovering-novel-algorithms-with-alphatensor/

It took 50 years to beat the Volker Strassen’s Strassen algorithm for matrix multiplication which marginally improves standard multiplication.

i.e. For multiplication of Two, 2×2 matrix, it takes 7 multiplications instead of 8 in normal multiplication. Also it is kind of recursive in the sense that it divides bigger matrix into smaller and then do multiplication,

so, that savings of multiplication will increase.

Deepmind’s Alpha Tensor has matched its performance with strassen for most and surpassed for many combinations.


Multiplying a 4 × 5 matrix by a 5 × 5 matrix, for example, the previous algorithm required 80 individual multiplications.

AlphaTensor needed only 76.


So, many times in my posts, I have focused on reducing the number of arithmatic operations (mult/div specially). be it one multiplier only, It will help so much.

One such techniques I have shared is multiplication of two complex numbers.

And many more techniques I have shared in the previous posts.

KUDOS TO DEEPMIND’s team that have developed Alpha Tensor.



#vlsi #vlsitraining  #verilog #verification   #vlsidesign  #fpga #functionalverification  #simulation



Discover more from Verilog Master

Subscribe to get the latest posts to your email.

One response to “P41 GENERIC KNOWLEDGE-2”

  1. thanks for the post

Leave a comment

Blog at WordPress.com.