4

Your questions phrasing is a bit confusing. I'll try to answer my interpretation of your confusion. I think the base of your confusion relays on the assumption that 2D convolution receives a 2 dimensional array as input. This is not true - The input of a 2D convolution layer is a 3 dimensional array. At the beginning you have an image which can be greyscale/...


3

Simple neural network as a structure doesn't have invariance across time scale deformation that's why it is impractical to apply it to recognize time series. To recognize time series usually a generic communication model is used (HMM). NN could be used together with HMM to classify individual frames of speech. In such HMM-ANN configuration audio is split on ...


2

The SIFT descriptor vector is a feature vector. "Descriptor vector" and "feature vector" are synonyms in this context. Most of the descriptions of SIFT I've seen use the phrase "descriptor vector", but occasionally they'll refer to it as a "feature vector" or refer it to as "SIFT features", perhaps to draw upon intuition from machine learning. SIFT works ...


2

First, let's talk about what a histogram of directions is. You can think of the image as a 2D discrete function of x and y: I(x,y). You can take partial derivatives of this function: Ix and Iy. So at each pixel you have the gradient, which is the vector (Ix(x,y), Iy(x,y)). If you compute the magnitude of the gradient, and look at it as an image, you will ...


2

As a preliminary answer, I can't imagine the extraction of AUs affecting their ability to detect emotions, since according to Do Deep Neural Networks Learn Facial Action Units When Doing Expression Recognition?, CNNs trained to extract emotions already correspond to AUs (called FAUs in the paper), as shown in the table below taken from their paper: ...


2

This is still a binary classification task. In the abstract, there are two ways to handle this: Most classifiers can output a predicted class and a confidence score (which indicates how confident the classifier is in its prediction). If you don't need a probability, you can use the confidence score. If you want to turn it into a probability, you can use ...


2

The primary reason for including a history of states is likely indeed the ko rule. Even if having a long history will often be redundant, it's unlikely to hurt either (except that it might take some computation time... but that doesn't appear to be a major concern throughout the entirety of the papers). I imagine that having a history of states can also ...


1

It looks like you are using RapidMiner documentation as a medium to study. If that is case, then the best answer to your question and probably to some other questions of yours should be, I believe, a recommendation of related textbook, tutorial or lecture notes. For the current question, I recommend you to read a tutorial on Principal Components Analysis by ...


1

"This doesn't seem to be discussed in either of the papers, " Yes, it is discussed in at least one of the papers. Here is an excerpt taken from Mastering the Game of Go without Human Knowledge - DeepMind by David Silver, Julian Schrittwieser, Karen Simonyan, et al. The emphasis on "repetitions are forbidden" is added by me. History features $X_t, Y_t$ ...


1

Yup, there are lots of them. For example, image segmentation through graph cuts, camera calibration, image morphing, image stitching, reconstruction of a 3D scene from multiple 2D images, optical flow algorithms, motion tracking, image restoration and inpainting, and probably many more.


1

There seems to be a misconception. You can use $k$-nearest neighbors with any number of dimensions. There is no prohibition on using it with a large number of features. And in some cases $k-NN will work well even in high dimensions (a large number of features); it all depends on the dataset, and to find out whether it will work well, you have to try it. ...


1

As you have noticed, the denominators in the definition of $x'_t,y'_t$ ultimately have no effect, as they will be normalized away. Nonetheless, they do have significance. This definition ensures that $x'_t,y'_t$ represent the derivatives. If you omitted the denominators in the definition of $x'_t,y'_t$, you'd get the same value for $\hat{x}'_t,\hat{x}'_t$,...


1

For the first part of your question, the significance is just to guarantee that the vector $(\hat{x}'_t,\hat{y}'_t)$ has length $1$. This means that it tells us in which direction $(x_t,y_t)$ is changing, but not by how much. For example, $(\hat{x}'_t,\hat{y}'_t)=(1,0)$ tells us that $x_t$ is increasing and $y_t$ is constant; $(\hat{x}'_t,\hat{...


1

In the simplified special case where each image $x$ has a single correct labelling $k=\kappa(x)$, this becomes just a regression problem: you want to find a function $q$ that minimizes the risk $$\mathbb{E}[||q(X)-\kappa(X)||^2],$$ where $X$ is a random variable distributed according to some distribution on faces. One way to solve this is to build up a ...


1

There is no single answer. A cascade is a very simple idea: it basically represents a bunch of classifiers, applied sequentially. You are free to decide how each individual cascade will work. You could design the cascade so that every classifier uses the same set of features. Or, you could design the cascade so that each classifier has a different subset ...


1

Yes. You want to look at the bag-of-words model and n-gram models. The bag-of-words model corresponds to the part of your characteristic vector: namely, the parts <A>, <B>, <C> of your characteristic vector, i.e., the frequencies of a single letter. These are used as features for the ML algorithm. n-gram models generalize this to a ...


1

I don't see how there should be any negative consequences of this. To see why, you can "vectorize" your feature matrix so that you still have scalars for each "new" feature. That is, instead of a matrix of n features each with k values, you have a vector of length n×k with scalar features. SVMs with a mixture of categorical/integral + real components is ...


1

Option 1: You can crop the images to the smallest sizes in all dimensions. However, blindly cropping images will cause you to lose important information, if you don't have a region of interest. For example, if you are focusing on faces, it is fine to define a ROI and crop around the face. On the other hand, for example if you are doing pure color-wise ...


Only top voted, non community-wiki answers of a minimum length are eligible