search

Lvchangze / snn

ICLR 2023, Spiking Convolutional Neural Networks for Text Classification
26 stars 3 forks source link

The problem of energy calculation #4

Open 3auxin opened 7 months ago

3auxin commented 7 months ago

Dear author,

In your paper, you mentioned a method for calculating SOP, but I encountered some difficulties when implementing it. I have carefully read through the description in the paper several times, but I still do not fully understand the details. I would like to inquire about how you implemented this SOP calculation method, and if there are any special considerations or techniques involved in the implementation process.

Lvchangze commented 7 months ago

Dear 3auxin: The energy estimation in this paper has been already outdated. Please refer to 《Attention spiking neural networks》Yao Man, TPAMI 2023. Besides, for every layer like conv or mlp, the SOPs = T fire rate FLOPs, where fire rate is the firing rate of the input in this layer. You can use Pytorch Hook to both judge which class current layer is and calculate fire rate of the input.

3auxin commented 7 months ago

Thank you for answering my question, I truly appreciate your assistance.