chan8972/enabling_spikebased_backpropagation

enabling_spikebased_backpropagation's Issues

Issue about model training on N-MNIST

Hi,

I'm trying to implement LeNet using the N-MNIST dataset but the accuracy does not increase as the training step growing. I load the N-MINST dataset by PySNN ( https://github.com/BasBuller/PySNN).

Can you please tell me how to train the model on N-MNIST?

Your help will be appreciated.
Thanks

Membrane potential and spike acitivity

hi,
could you plz tell me how to record spike activities?

Estimating Energy consumption

Hi,

I wanna calculate the total energy consumption of the ANN and SNN model.

I've read your article that you calculated the total energy consumption for both ANN and SNN models, which I got it.

For the SNN model, total energy consumption is calculated as (# of neurons in a particular layer * # of synaptic connections* average spiking activity).

I didn't exactly understand, how did you calculate the total energy for VGG and ResNet for SNN and ANN models which you've reported in your article? I mean the ratios?

For instance, I've made 4 layer SNN and ANN model for the MNIST dataset then how could I calculate the total energy?

I mean, how I know the total number of spikes in one inference image?

Do you have any code for that, so that I can calculate or Can you please give me an example??

Your help will be highly appreciated.

Thanks

Issues in training from scratch

Hi,

When I am trying to train from scratch, I get such an error :

"Traceback (most recent call last):
File "cifar10_ResNet11.py", line 704, in
main()
File "cifar10_ResNet11.py", line 143, in main
train(train_loader, model, criterion, criterion_en, optimizer, epoch, time_steps=100, leak=0.99)
File "cifar10_ResNet11.py", line 206, in train
output, Total_11_output, LF_11_output, Total_21_output, LF_21_output, Total_22_output, LF_22_output, Total_31_output, LF_31_output, Total_32_output, LF_32_output, Total_41_output, LF_41_output, Total_42_output, LF_42_output, Total_51_output, LF_51_output, Total_52_output, LF_52_output, Total_f0_output, LF_f0_output, out11_temp, out21_temp, out22_temp, out31_temp, out32_temp, out41_temp, out42_temp, out51_temp, out52_temp, outf0_temp = model(input_var, steps=time_steps, l=leak)
File "/home/ubuntu/anaconda3/envs/pytorch_p27/lib/python2.7/site-packages/torch/nn/modules/module.py", line 532, in call
result = self.forward(*input, **kwargs)
File "cifar10_ResNet11.py", line 570, in forward
mem_11, out = LIF_sNeuron(mem_11, self.cnn11.threshold, l, i)
File "cifar10_ResNet11.py", line 411, in LIF_sNeuron
out = SpikingNN()(ex_membrane)
TypeError: forward() takes exactly 2 arguments (1 given)".

Can you please check from your side?

Forward Function

Hi,

I am referring to your code. I have some confusion related to forward function.

I'm trying to implement VGG11 using the CIFAR10 dataset but got some dimension errors.

Can you please explain how did you make the dimensions inside forward function like this?

torch.zeros(batch_size, 64, 32, 32, device=device))

When I tried this using VGG11 , got this error?

It means my dimensions are not correct.

Below is the snippet of my forward function:

Issue about Poisson encoding method

I notice the implementation of Poisson-distributed spike as below

rand_num = Variable(torch.rand(input.size(0), input.size(1), input.size(2), input.size(3)).cuda())
Poisson_d_input = (torch.abs(input) > rand_num).type(torch.cuda.FloatTensor)
Poisson_d_input = torch.mul(Poisson_d_input, torch.sign(input))

which means that the encoded spikes are triple-valued {0, ±1}. It's a little different from the well-known encoding style with binary value {0, 1}.

I wonder why you choose this way of Poisson encoding.