Hi, First of all great work.
Wanted to know if the code for convolution LGN is going to be publicly available anytime soon?
Also, I was quite curious about the adder tree structure that mimics GroupSum technique in software.
Is it a balanced pipelined adder tree structure? Because I feel in terms of latency, the neural network structure is optimised, but the bottleneck might be adder tree structure.
Hi, First of all great work.
Wanted to know if the code for convolution LGN is going to be publicly available anytime soon?
Also, I was quite curious about the adder tree structure that mimics GroupSum technique in software.
Is it a balanced pipelined adder tree structure? Because I feel in terms of latency, the neural network structure is optimised, but the bottleneck might be adder tree structure.