Methods and principles of building "computer systems that automatically improve with experience."

learn more… | top users | synonyms (1)

0
votes
0answers
1 view

Will normalizing training and testing data separately cause under/overfitting?

Suppose I have training and testing data and I want to train a classifier (e.g. SVM). Typically, features are normalized before classification to ensure some features aren't weighted more heavily than ...
0
votes
0answers
4 views

Prove Reccurrent Neural Network can exhibit oscillatory behavior

I understand how recurrent neural networks work, however I'm trying to build a deep intuitive understanding of their behavior which is difficult for me because they exhibit such complex behaviors. ...
0
votes
0answers
4 views

Interpreting crfsuite output model for numerical features

I am using crfsuite-python to implement a linear chain CRF in which I would like to use numerical features rather than strings as is the case with the standard CRF application parts of speech tagging. ...
1
vote
0answers
7 views

Ideal statistical or machine learning technique to predict streamflow from snow albedo (highly cross-correlated)?

I'm trying to build a model that can predict streamflow for an alpine (snowmelt-fed) watershed using snow albedo (roughly, the energy reflectance of the snow) data. Albedo controls the melt of the ...
1
vote
1answer
8 views

Need a little help understanding K-means++ seeding

I have been working on a project that involves using K-means clustering for generating adaptive palettes from images. I understand the general process of K-means clustering, and I understand the ...
0
votes
0answers
14 views

R: Finding relationships between 2 variables to determine any patterns in data

I am working on finding relationships/patterns between 2 variables (Type_A, Type_B). ...
0
votes
2answers
23 views

Generative algorithms

If I understand slide 4 correctly, the idea here is that in order to compute $p(y|x)$ we can use the fact that $$p(y|x) = \frac{p(x|y)p(y)}{p(x)}.$$ Then $p(x|y)$ and $p(y)$ are calculated using our ...
3
votes
1answer
12 views

do CART trees capture interactions among predictors?

This paper claims that in CART, because a binary split is performed on a single covariate at each step, all splits are orthogonal and therefore interactions among covariates are not considered. ...
1
vote
0answers
17 views

Are Random Forests and Boosting parametric or non-parametric?

From this excellent paper by Breiman, we can seize all the difference between traditional statistical models (e.g., linear regression) and machine learning algorithms (e.g., Bagging, Random Forests, ...
0
votes
0answers
8 views

Variable representations for faster learning convergence

My notes on machine learning state that transforming a classification problem from 2 classes, class A = 0, and class B = 1, to class A = $(1,0)$, and class B = $(0,1)$ leads to faster convergence in ...
1
vote
0answers
13 views

How could these two simple Bayesian algorithms be explained, simply? [on hold]

count(this token in class) + 1 / count(all tokens in class) + count( all tokens ) and ...
2
votes
1answer
18 views

How to make use of less data of a particular class for better modeling?

My question may be dump. But I am all confused to start with. I am having a set of dataset, say 9000 rows, with some features. Around 8000 belongs to class "1" and 1000 goes to class "0". So, if I am ...
0
votes
0answers
23 views

In machine learning, may I train correctly a neural network with input real data and output validation Boolean data?

I have a matrix made of ~ 100 rows and 12 columns. Each entry contains a real value. The first 6 columns refer to a particular concept (firstClass), the following 6 to another one (secondClass), and ...
0
votes
0answers
12 views

What is the honesty condition for regression trees?

I have a question pertaining to Stefan Wager's "Asymptotic Theory for Random Forests": http://arxiv.org/pdf/1405.0352v1.pdf Wager first states that trees are "fully grown in the sense given training ...
0
votes
0answers
10 views

RBM hidden units becoming correlated

I am trying to train an RBM with 8 hidden binary units and 40 visible ReLUs. At first, I had issues with binary units becoming stuck due to the weight saturating, but I got rid of that problem by ...
-1
votes
1answer
20 views

What are the benefits for semi-supervised learning over unsupervised clustering? Or any limitations?

I have another question about semi-supervised learning vs unsupervised clustering, what are the benefits and limitations? I have got some data with labels and some without labels. I performed ...
1
vote
0answers
12 views

Semi-supervised learning vs supervised learning, what are the benefits and limitations?

Just wondering if any previous work compared semi-supervised learning vs supervised learning? Currently, I have got both datasets with and without labeling. And therefore, it is intuitive for me to ...
0
votes
1answer
12 views

Determine confidence in a CART model with factor (2 levels) response variable (using rpart)

I use the package rpartto model a classification/regression tree. I have the variables $x,y,s$ where $x$ is in $\{-1,1\}$, y is continuous in $[0,1]$ and $s$ is a ...
2
votes
0answers
25 views

Reinforcement Learning in Industry [on hold]

This is my first post here I would like to start with a rather general topic of discussion. I have studied Reinforcement Learning during the university years and although I find it rather fascinating ...
1
vote
1answer
24 views

Expectation of squared error

In machine learning, we let $X$ be a real-valued input vector and $Y$ be a real number output, with joint distribution $P(X,Y)$. We are looking for a function $f(X)$ for predicting $Y$ given the ...
0
votes
0answers
21 views
0
votes
1answer
24 views

Online gradient descent for strongly convex function

Given that our loss function is $\alpha$ strongly convex function which means $\mbox(\nabla f(\mathbf{x})-\nabla f(\mathbf{y}))^{T}(\mathbf{x}-\mathbf{y})\geq \alpha||\mathbf{x}-\mathbf{y}||_{2}^{2} ...
3
votes
1answer
28 views

How is the confusion matrix reported from K-fold cross-validation?

Suppose I do K-fold cross-validation with K=10 folds. There will be one confusion matrix for each fold. When reporting the results, should I calculate what is the average confusion matrix, or just sum ...
0
votes
0answers
6 views

Gaussian classifier: if two gaussians have equal variance is it possible for them to produce a non-linear decision boundary?

I have been playing with this a bit and I don't believe they can. However, I am very new to machine learning and my maths isn't strong enough to be certain.
0
votes
1answer
28 views

Is this feature redundant?

Say I have a data set, and there's one feature that divides the set into roughly two halves, labeling one half A, and the other half B. Now I have another feature, it labels all instances that were ...
1
vote
1answer
34 views

Neural Network System Identification

I am trying to implement a Neural Network to identify a Nonlinear System. I have implemented a very simple system in simulink and on the basis of examples of its input and output I would like to have ...
2
votes
0answers
27 views

Gaussian Mixture and K-Means ?! a big challenge?

This is taken from Tom. Mitche Material as Old-Exam. I think the (2) is true and not (3). Who can verify me?
0
votes
0answers
7 views

Train model based on correlations

I have a dataset of all trains in my country for a period in time, in a MySQL database. The form of this data is the following: ...
0
votes
0answers
17 views

Decision Tree in test, a Wrong Problems? [on hold]

I took a test two days ago. one of our question is as follows: decision tree with depth 2 is constructed for two binary feature. hypothesis spase that can be shown with the following tree has ...
0
votes
0answers
32 views

Does Dynamic Bayesian Network have to be symmetric?

I want to create a Dynamic Bayesian Network with 2 time slices, each with 12 nodes. This is the network I made: Some people in my research group said that this is not DBN because its inter-slice ...
0
votes
0answers
13 views

N-gram learning vs stochastic learning

I'm interested in comparing the differences in learning in n-grams and gradient-based learning (in my case with neural networks), particularly in the context of language modelling with the two classes ...
0
votes
0answers
16 views

K-cross validation and Naive Bayes

I am doing an exercise of machine learning, and I have built a Gaussian Naive Bayes classifier (i.e., I have defined values of mean and standard deviation) using scikit-learn. Now I am supposed to ...
0
votes
1answer
41 views

Neural network & Bayesian in this machine learning algorithm

I am new to machine learning etc and found this comprehensive algorithm: http://scikit-learn.org/stable/tutorial/machine_learning_map/ . However, I am not able to make out any reference to neural ...
1
vote
0answers
21 views

An example for a finite hypothesis class which is not PAC learnable?

Finite hypothesis class with bounded loss function are PAC learnable. Are there examples for finite hypothesis classes in the case of unbounded loss function, which aren't PAC learnable?
0
votes
0answers
44 views

R:Text Analysis and Classify as type

I am new to R and Analysis, I have content set (emails) that are stored as csv file that is of more than 1000 rows(more than one email content in a row) , these are been imported to R and have been ...
0
votes
0answers
12 views

R: Caret Package - Regression with (strong) confidence?

I am fairly new to using prebuilt machine learning packages in R. I am looking at the following problem. I have a long feature set, very small training set, and a large test set. The goal is to ...
0
votes
0answers
6 views

Denoising Autoencoders weights at test time

When using masking noise on Denoising Autoencoders,Should weights be increased at test time proportional to the masking rate as in Dropout?
0
votes
0answers
20 views

PAC learning model definition

The probably approximately correct (PAC) learning model definition is: A concept class $C$ is said to be PAC-learnable if there exists an algorithm $A$ and a polynomial function $poly(·,·,·,·)$ such ...
1
vote
0answers
34 views

Support Vector Machines vs KNN

It was my understanding that in a separable case, SVMs produce the best separation possible and therefore will always produce the same or a better classification rate compared with say, 1NN, ...
0
votes
0answers
6 views

Updating model parameters online on test data

I learn the parameters of a temporal model (in my case, an RTRBM) on some training sequences using mini-batch gradient descent. Let's say now that I am updating my model online after every prediction ...
0
votes
0answers
16 views

implementation of Poisson regression [duplicate]

I am trying to work with Poisson regression. I came across this video which is very helpful - https://www.youtube.com/watch?v=HntUY8SsYZg. In the video one of parameters (Race) is categorical and ...
-1
votes
0answers
13 views

embed histogram into larger feature vector

in my problem I have feature sets extracted by different methods and I want to put them into one single feature vector. However, one of these feature sets is a histogram which I don't know what's the ...
0
votes
0answers
13 views

What is inference, training and testing in Undirected Graphical Model? [on hold]

I have a Undirected Graphical Model (UGM) - $ \sum_i w_i\phi_i $ . What is the inference and training here? Suppose I have a train data and test data how do I train and test using this data and UGM? ...
0
votes
0answers
17 views

predict for rpart model [closed]

I do cross validation for doing rpart model, exactly I do leave one out(LOO)(one row fro testing teh model and the others for learning teh model) so the testing set will consist of one row for each ...
0
votes
0answers
13 views

How to determine number of feature maps in the convolution layer of a CNN?

How do we know how many feature maps is needed in the convolution layer? Other steps is clear to me except that convolution steps.
1
vote
0answers
16 views

hinge loss vs logistic loss advantages and disadvantages/limitations

Hinge loss can be defined using $\text{max}(0, 1-y_i\mathbf{w}^T\mathbf{x}_i)$ and the log loss can be defined as $\text{log}(1 + \exp(-y_i\mathbf{w}^T\mathbf{x}_i))$ I have the following questions: ...
0
votes
0answers
28 views

SVM output to probabilistic affiliation

How can I convert the svm output for multiple class classification(one vs one approach) to probabilistic values? Meaning that I want to have a probability for a tested element to be in each available ...
-1
votes
0answers
33 views

Nearest algorithm according to which the humans analyze the data [closed]

Which Algorithm analyze the data just like the people does? Nearest algorithm according to which the humans analyze the data Can I say that the people group the data similar to the s.link algorithm ...
1
vote
1answer
27 views

A question about SVM kernels

this is a very basic question about SVM. I was using SVMs that are provided in the scikit for some problems, and noted that they are quite slow for big datasets. I then learned more about the ...
2
votes
2answers
54 views

Using Neural Net weights as input to another classifer

Is there anyway to use the weights from a neural net hidden layer as input to another classifier, say a random forest? Of course this is trivial for the training data but how to score new data? Are ...