verne.freenode.net changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
marcosirc has quit [Quit: WeeChat 1.4]
tham has joined #mlpack
mentekid has joined #mlpack
nilay has joined #mlpack
tham has quit [Quit: Page closed]
< nilay> zoq: Hi
< nilay> if i need an intermediate output, (when we do concatenation step) what layer should be used
< zoq> Hello nilay, do you mean the concatenation with the auxiliary classifier?
< nilay> well i think it could be with anything, so we will have 4 separate nets in inception layer one doing 1x1 conv, one 3x3, one 5x5 and one pooling, each of these will have some outputs, which we concatenate and this is the output of inception layer, right?
< zoq> Ah, I see, you mean the filter concatenation in the inception layer. I think, since the output has the same dimension for all layers inside the inception layer, it's sufficient, to just accumulate the output in the inception layer. What do you think
< zoq> ?
< nilay> yeah so when i make a net as net(modules, someOutputLayer) what should the type of someOutputLayer be?
< zoq> I'm not sure, you have to create a net inside the inception layer, you could just use the layer independently and call the forward function, to get the output.
< zoq> Something like:
< zoq> Conv(1x1).Forward(input, output1x1)
< zoq> Conv(3x3).Forward(output1x1, output3x3)
< zoq> Conv(5x5).Forward(output1x1, output5x5)
< zoq> output1x1 + output3x3 + output5x5
< nilay> Conv(1x1) is a ConvLayer<>?
< zoq> yes
< zoq> In the Constructor you could create all layer needed; in the Forward pass you call the Forward function from each layer, with the right input and at the end you accumulate the output from each layer.
< nilay> ok and i do similarly for Backward?
< zoq> yes, right
< nilay> ok , can i do something about cnn.hpp and cnn_impl.hpp, i couldon't understand the code there much
< nilay> couldn't*
< nilay> too much template syntax knowledge
< zoq> hm, maybe it helps if I'll send you some further informations?
mentekid has quit [Ping timeout: 250 seconds]
< nilay> yeah, that would be great
< zoq> okay, sure, no problem, I can do that later.
< nilay> ok
kwikadi has joined #mlpack
nilay has quit [Ping timeout: 250 seconds]
nilay has joined #mlpack
mentekid has joined #mlpack
mentekid has quit [Ping timeout: 244 seconds]
mentekid has joined #mlpack
nilay has quit [Ping timeout: 250 seconds]
mentekid has quit [Ping timeout: 264 seconds]