Trjtdtk

Does Gita support doctrine of eternal cycle of birth and death for evil people?

Rivers without rain

Can SQL Server create collisions in system generated constraint names?

Reducing vertical space in stackrel

How does a program know if stdout is connected to a terminal or a pipe?

How did Captain America manage to do this?

Is there really no use for MD5 anymore?

How do I reattach a shelf to the wall when it ripped out of the wall?

Examples of subgroups where it's nontrivial to show closure under multiplication?

Do I have an "anti-research" personality?

Why is it that the natural deduction method can't test for invalidity?

What does the "ep" capability mean?

How much cash can I safely carry into the USA and avoid civil forfeiture?

Don’t seats that recline flat defeat the purpose of having seatbelts?

Why was Germany not as successful as other Europeans in establishing overseas colonies?

How can the Zone of Truth spell be defeated without the caster knowing?

How to reduce LED flash rate (frequency)

Was there a shared-world project before "Thieves World"?

How do I deal with a coworker that keeps asking to make small superficial changes to a report, and it is seriously triggering my anxiety?

Please, smoke with good manners

How to verbalise code in Mathematica?

How to make a pipeline wait for end-of-file or stop after an error?

Is there a way to get a compiler for the original B programming language?

Examples of non trivial equivalence relations , I mean equivalence relations without the expression " same ... as" in their definition?

Paper about AE-CNN is unclear. Deriving layers of dense blocks?

Stuck on deconvolution in Theano and TensorFlowHow does a convolutional ply differ from an ordinary convolutional network?Do Convolution Layers in a CNN Treat the Previous Layer Outputs as Channels?How to create a multiple layer perceptron with layers of specific sizes in keras?Should there be a flat layer in between the conv layers and dense layer in YOLO?Several fundamental questions about CNNTraining Accuracy stuck in KerasCombining spatial input with a label as input for CNN using KerasDetermining size of FC layer after Conv layer in PyTorchReconstructing input image from layers of a CNN

1

$begingroup$

I am implementing the algorithm called Automatically Evolving CNN (AE-CNN).

Some things aren't specified which makes it a bit hard to understand what the paper actually means to say. In the chapter 3.2 Encoding Strategy it says:

...Note that the number of convolutional layers
in a DB is known because it can be derived by the spatial
sizes of input and output as well as k. ...

By DB it means a single dense block from the DenseNet algorithm. K means the K parameter for the dense block. This is the growth rate (how many layers each conv layer adds). Later on, the paper restricts the allowed values for K to just 12, 20, and 40.

I have two questions about this statement:

1. Given the shape of the input- and output tensors and a growth rate, how would one derive the number of layers?

The block cannot change the width and height of the feature maps (except for not padding but this is not the case) nor can it change the depth of a tensor. The depth of the output tensor can only be equal to the chosen K (I believe that the other feature maps must be ignored after the last conv layer).

2. The statement is only talking about single DBs but, later on, the paper says that it groups multiple DBs into units (DBU) and that the input- and output tensor shapes are only specified per units. This is contradictory. How should I interpret this? Does the statement makes sense if it would talk about DBUs instead?

Note: The paper isn't specific about it but I interpreted the paper to say that the depth of either the input tensor or output tensor (but not both) should be evolved.

Edit: A follow-up question: would it make sense to evolve this parameter (the number of layers in a DB or DBU)?

machine-learning cnn convnet evolutionary-algorithms

share|improve this question

edited Apr 7 at 18:54

Jupiter

asked Apr 7 at 18:06

JupiterJupiter

1133

$endgroup$

add a comment | 

1

$begingroup$

I am implementing the algorithm called Automatically Evolving CNN (AE-CNN).

Some things aren't specified which makes it a bit hard to understand what the paper actually means to say. In the chapter 3.2 Encoding Strategy it says:

...Note that the number of convolutional layers
in a DB is known because it can be derived by the spatial
sizes of input and output as well as k. ...

By DB it means a single dense block from the DenseNet algorithm. K means the K parameter for the dense block. This is the growth rate (how many layers each conv layer adds). Later on, the paper restricts the allowed values for K to just 12, 20, and 40.

I have two questions about this statement:

1. Given the shape of the input- and output tensors and a growth rate, how would one derive the number of layers?

The block cannot change the width and height of the feature maps (except for not padding but this is not the case) nor can it change the depth of a tensor. The depth of the output tensor can only be equal to the chosen K (I believe that the other feature maps must be ignored after the last conv layer).

2. The statement is only talking about single DBs but, later on, the paper says that it groups multiple DBs into units (DBU) and that the input- and output tensor shapes are only specified per units. This is contradictory. How should I interpret this? Does the statement makes sense if it would talk about DBUs instead?

Note: The paper isn't specific about it but I interpreted the paper to say that the depth of either the input tensor or output tensor (but not both) should be evolved.

Edit: A follow-up question: would it make sense to evolve this parameter (the number of layers in a DB or DBU)?

machine-learning cnn convnet evolutionary-algorithms

share|improve this question

edited Apr 7 at 18:54

Jupiter

asked Apr 7 at 18:06

JupiterJupiter

1133

$endgroup$

add a comment | 

$begingroup$

I am implementing the algorithm called Automatically Evolving CNN (AE-CNN).

Some things aren't specified which makes it a bit hard to understand what the paper actually means to say. In the chapter 3.2 Encoding Strategy it says:

...Note that the number of convolutional layers
in a DB is known because it can be derived by the spatial
sizes of input and output as well as k. ...

By DB it means a single dense block from the DenseNet algorithm. K means the K parameter for the dense block. This is the growth rate (how many layers each conv layer adds). Later on, the paper restricts the allowed values for K to just 12, 20, and 40.

I have two questions about this statement:

1. Given the shape of the input- and output tensors and a growth rate, how would one derive the number of layers?

The block cannot change the width and height of the feature maps (except for not padding but this is not the case) nor can it change the depth of a tensor. The depth of the output tensor can only be equal to the chosen K (I believe that the other feature maps must be ignored after the last conv layer).

2. The statement is only talking about single DBs but, later on, the paper says that it groups multiple DBs into units (DBU) and that the input- and output tensor shapes are only specified per units. This is contradictory. How should I interpret this? Does the statement makes sense if it would talk about DBUs instead?

Note: The paper isn't specific about it but I interpreted the paper to say that the depth of either the input tensor or output tensor (but not both) should be evolved.

Edit: A follow-up question: would it make sense to evolve this parameter (the number of layers in a DB or DBU)?

machine-learning cnn convnet evolutionary-algorithms

share|improve this question

edited Apr 7 at 18:54

Jupiter

asked Apr 7 at 18:06

JupiterJupiter

1133

$endgroup$

share|improve this question

edited Apr 7 at 18:54

Jupiter

asked Apr 7 at 18:06

JupiterJupiter

1133

share|improve this question

edited Apr 7 at 18:54

Jupiter

asked Apr 7 at 18:06

JupiterJupiter

1133

asked Apr 7 at 18:06

JupiterJupiter

1133

asked Apr 7 at 18:06

JupiterJupiter

1133