Trjtdtk

Stars Make Stars

Estimated State payment too big --> money back; + 2018 Tax Reform

What is the electric potential inside a point charge?

Cold is to Refrigerator as warm is to?

What did Darwin mean by 'squib' here?

What are the performance impacts of 'functional' Rust?

Is it possible to ask for a hotel room without minibar/extra services?

Mortgage adviser recommends a longer term than necessary combined with overpayments

Cauchy Sequence Characterized only By Directly Neighbouring Sequence Members

Classification of bundles, Postnikov towers, obstruction theory, local coefficients

How do you clear the ApexPages.getMessages() collection in a test?

Need a suitable toxic chemical for a murder plot in my novel

What do I do if technical issues prevent me from filing my return on time?

How does modal jazz use chord progressions?

Fishing simulator

What LEGO pieces have "real-world" functionality?

Slither Like a Snake

The following signatures were invalid: EXPKEYSIG 1397BC53640DB551

What's the point in a preamp?

Can I add database to AWS RDS MySQL without creating new instance?

Can smartphones with the same camera sensor have different image quality?

Jazz greats knew nothing of modes. Why are they used to improvise on standards?

What would be Julian Assange's expected punishment, on the current English criminal law?

What is the largest species of polychaete?

understanding densenets

0

$begingroup$

I have been reading the paper on DenseNet (https://arxiv.org/pdf/1608.06993) and I have a confusion (and this extends to Resnets) and other variants for creating deep networks:

• Why do dense connections generate less redundancy in the learnt features? I understand that the input is shared but what is the mechanism that stops the layers learning the same thing? I somehow do not see how this can ensure lack of redundancy?




• Why do the authors say preactivation is more efficient than the usual post activation in a dense block? Is there a good theoretical or even intuitive reasoning for this?




• How do dense connections act as regularization? I thought having more sense connections should lead to more chances of overfitting but that is not what the authors claim.

neural-network

share|improve this question

asked Apr 2 at 8:36

LucaLuca

1386

$endgroup$

add a comment | 

0

$begingroup$

I have been reading the paper on DenseNet (https://arxiv.org/pdf/1608.06993) and I have a confusion (and this extends to Resnets) and other variants for creating deep networks:

• Why do dense connections generate less redundancy in the learnt features? I understand that the input is shared but what is the mechanism that stops the layers learning the same thing? I somehow do not see how this can ensure lack of redundancy?




• Why do the authors say preactivation is more efficient than the usual post activation in a dense block? Is there a good theoretical or even intuitive reasoning for this?




• How do dense connections act as regularization? I thought having more sense connections should lead to more chances of overfitting but that is not what the authors claim.

neural-network

share|improve this question

asked Apr 2 at 8:36

LucaLuca

1386

$endgroup$

add a comment | 

$begingroup$

I have been reading the paper on DenseNet (https://arxiv.org/pdf/1608.06993) and I have a confusion (and this extends to Resnets) and other variants for creating deep networks:

• Why do dense connections generate less redundancy in the learnt features? I understand that the input is shared but what is the mechanism that stops the layers learning the same thing? I somehow do not see how this can ensure lack of redundancy?




• Why do the authors say preactivation is more efficient than the usual post activation in a dense block? Is there a good theoretical or even intuitive reasoning for this?




• How do dense connections act as regularization? I thought having more sense connections should lead to more chances of overfitting but that is not what the authors claim.

neural-network

share|improve this question

asked Apr 2 at 8:36

LucaLuca

1386

$endgroup$

share|improve this question

asked Apr 2 at 8:36

LucaLuca

1386

share|improve this question

asked Apr 2 at 8:36

LucaLuca

1386

asked Apr 2 at 8:36

LucaLuca

1386

asked Apr 2 at 8:36

LucaLuca

1386