E_net4

Did the student take any programming courses during the first year? I would see it as something essential for a student without a CS/CE background doing computational engineering.

While the supposed experts in machine learning should stick to this site or CV, a fair bunch of Stack Overflow users are accepting pure machine learning questions in there.

I've seen NN debugging questions all over the three sites (SO/DS/CV). I sometimes vote to lose those on SO because they're either incomplete or do not relate enough to software development. They often attract opinionated answers too ("I don't know, but here are some tips: ...", "try this instead: ...").

So, has anyone thought through this matter?

That is fine!

pong

It's up to you. I really don't know what you are trying to do, but you should consider using the 4096-dimensional vector first. It's the usual methodology: hypothesize, experiment, measure, repeat.

This depends on too many factors, such as the dataset and the trained network. Empirically, a single similar element might not be significant, but a small Euclidean distance between two vectors might.

The presence of the same value in the same vector index may convey one particular similarity, but that may not be trivial. In general, a good feature representation means that similar images will also have similar creatures under that representation.

Not to a trivial kind of feature, but yes. See this paper, it's an interesting read on this topic: arxiv.org/abs/1311.2901

2) on the other hand, the relation between a specific value in the vector and the effectively captured visual features is not trivial (this was added in the previous edit). However, there are known techniques to visualize how neural networks work.

1) the order matters. They represent distinct features. Shuffling the vector would result in a completely different feature vector.

Well, yes. But I still wonder why this discussion is needed.

I have updated the answer. On the other hand, it seems that you could use some further reading for the second question, which is a bit too open-ended to be covered here. There's the question Is there any difference between feature extraction and feature learning? and this paper on representation learning Representation Learning: A Review and New Perspectives is a strong overview on the subject of feature learning, where deep learning is often employed.

The short answer to that is yes. The long answer is that there may be many feasible ways to feed these features to an RNN. We could indeed reduce the dimensionality of the visual features, either by extracting neural codes from a new pre-trained network with a 100-dimensional layer, or just making an autoencoder or RBM. One could probably also feed the RNN with partitions of the features. This could make a new question of its own, if sufficiently focused to make it answerable.

Obtaining a bag of words with these features is usually unnecessary, because the vector of 4096 values already represents the entire image. You may picture the neural network as a complex composition of convolutions doing specific visual detections (edge, colour, ...) and adding the resulting activations together with some non-linearities in between. However, it is possible to reduce the dimensionality of these features in a number of ways. In Amazon's paper, they created a K-dimensional embedding of these features.

Oo, we have a veteran site grower. :)

Greetings, everyone. Looking forward to see some great content. :)

Nesta minha pergunta, estão duas repostas candidatas à recompensa. Convido-vos a votar nestas, porque vou atribuir a bounty à que tiver mais votos.

Saudações. :-)

Quanto à tal pergunta do facto e fato, acabei por fazer uma resposta também. tl;dr Está no VOP

Pois... é da fonte. x)

Portanto, achei piada. :P

Pi ~= 3.14159265358979323846

Nope.

3141 :P

@JorgeB. your rep is approx. π * 1000 right now. Impressive!

Alicerces.

Entretanto, lá tratei de responder à minha própria dúvida. A ver se alguém comenta no assunto.

De todos os modos, é uma boa pergunta.

Essa tag está bem colocada. Não consigo pensar em mais. :|

Precisamente.

Já imaginaram o que seria ter perguntas de tradução para qualquer outra língua?

Tive algum receio nos primeiros dias dessa pergunta.

Sure, I'm happy to contribute to the site and make a stand in Meta from time to time, but being a mod is one step I probably shouldn't take.

Both, I guess.

I'm not urged to do that at all. :P

Good point.

"Now"?

Sei que sim. Mas já ajuda dar o link, a resposta deverá ficar dentro dos moldes da outra, salvo um aspeto ou outro.

São os de ouros ♦.

(sqn)

Finos para todos!

:D

Yay! Party!

Está um dia fantástico! Chuva no verão!

@CiganoMorrisonMendez Só apontei.

Malhação, treinamento... Isso é tudo BR.

Mas é assim, afirma-se que o presente do conjuntivo é usado para expressar uma ação ou situação eventual (que poderá acontecer ou não, embora tipicamente desejada) no futuro.

Admito que fui refletindo neste assunto, mas até agora não consegui conceber uma resposta irrepreensível. Tanto mais que tenho certas prioridades até ao fim do mês, não posso perder muito tempo.