I'm doing a recognition problem (faces) and trying to reduce the problem size. I originally began with training data in a feature-wise coordinate system in 120 dimensions, but through PCA I found a better PC-wise coordinate system needing only 20 dimensions while still conveying 95% of the data.
I began thinking that recognition by definition is a problem of classification. Points in n-space belonging to the same object/face/whatever would cluster. To take an example, if 5 instances of the same individual are in the training data, they would cluster and the mid-point of that cluster could be numerically defined using k-means.
I have 100,000 observations, each person is represented by 5-10 headshots, this means instead of comparing a novel input to 100,000 points in my 20-space, I could instead compare to 10,000-20,000 centroids. Can k-means be used like this or have I misinterpreted? k is obviously undefined but I've been reading up on ways to find optimal k.
My specific recognition problem doesn't use neural nets but rather simple arithmetic euclidean distances between points.
I'm playing around with Neural Networks trying to understand the best practices for designing their architecture based on the kind of problem you need to solve.
I generated a very simple data set composed of a single convex region as you can see below:
Everything works fine when I use an architecture with L = 1, or L = 2 hidden layers (plus the output layer), but as soon as I add a third hidden layer (L = 3) my performance drops down to slightly better than chance.
I know that the more complexity you add to a network (number of weights and parameters to learn) the more you tend to go towards over-fitting your data, but I believe this is not the nature of my problem for two reasons:
my performance on the Training set is also around 60% (whereas over-fitting typically means you have a very low training error and high test error),
and I have a very large amount of data examples (don't look at the figure that's only a toy figure I uplaoded).
Can anybody help me understand why adding an extra hidden layer gives
me this drop in performances on such a simple task?
Here is an image of my performance as a function of the number of layers used:
ADDED PART DUE TO COMMENTS:
I am using a sigmoid functions assuming values between 0 and 1, L(s) = 1 / 1 + exp(-s)
I am using early stopping (after 40000 iterations of backprop) as a criteria to stop the learning. I know it is not the best way to stop but I thought that it would ok for such a simple classification task, if you believe this is the main reason I'm not converging I I might implement some better criteria.
At least on the surface of it, this appears to be a case of the so-called "vanishing gradient" problem.
Activation functions
Your neurons activate according to the logistic sigmoid function, f(x) = 1 / (1 + e^-x) :
This activation function is used frequently because it has several nice properties. One of these nice properties is that the derivative of f(x) is expressible computationally using the value of the function itself, as f'(x) = f(x)(1 - f(x)). This function has a nonzero value for x near zero, but quickly goes to zero as |x| gets large :
Gradient descent
In a feedforward neural network with logistic activations, the error is typically propagated backwards through the network using the first derivative as a learning signal. The usual update for a weight in your network is proportional to the error attributable to that weight times the current weight value times the derivative of the logistic function.
delta_w(w) ~= w * f'(err(w)) * err(w)
As the product of three potentially very small values, the first derivative in such networks can become small very rapidly if the weights in the network fall outside the "middle" regime of the logistic function's derivative. In addition, this rapidly vanishing derivative becomes exacerbated by adding more layers, because the error in a layer gets "split up" and partitioned out to each unit in the layer. This, in turn, further reduces the gradient in layers below that.
In networks with more than, say, two hidden layers, this can become a serious problem for training the network, since the first-order gradient information will lead you to believe that the weights cannot usefully change.
However, there are some solutions that can help ! The ones I can think of involve changing your learning method to use something more sophisticated than first-order gradient descent, generally incorporating some second-order derivative information.
Momentum
The simplest solution to approximate using some second-order information is to include a momentum term in your network parameter updates. Instead of updating parameters using :
w_new = w_old - learning_rate * delta_w(w_old)
incorporate a momentum term :
w_dir_new = mu * w_dir_old - learning_rate * delta_w(w_old)
w_new = w_old + w_dir_new
Intuitively, you want to use information from past derivatives to help determine whether you want to follow the new derivative entirely (which you can do by setting mu = 0), or to keep going in the direction you were heading on the previous update, tempered by the new gradient information (by setting mu > 0).
You can actually get even better than this by using "Nesterov's Accelerated Gradient" :
w_dir_new = mu * w_dir_old - learning_rate * delta_w(w_old + mu * w_dir_old)
w_new = w_old + w_dir_new
I think the idea here is that instead of computing the derivative at the "old" parameter value w, compute it at what would be the "new" setting for w if you went ahead and moved there according to a standard momentum term. Read more in a neural-networks context here (PDF).
Hessian-Free
The textbook way to incorporate second-order gradient information into your neural network training algorithm is to use Newton's Method to compute the first and second order derivatives of your objective function with respect to the parameters. However, the second order derivative, called the Hessian matrix, is often extremely large and prohibitively expensive to compute.
Instead of computing the entire Hessian, some clever research in the past few years has indicated a way to compute just the values of the Hessian in a particular search direction. You can then use this process to identify a better parameter update than just the first-order gradient.
You can learn more about this by reading through a research paper (PDF) or looking at a sample implementation.
Others
There are many other optimization methods that could be useful for this task -- conjugate gradient (PDF -- definitely worth a read), Levenberg-Marquardt (PDF), L-BFGS -- but from what I've seen in the research literature, momentum and Hessian-free methods seem to be the most common ones.
Because the number of iterations of training required for convergence increases as you add complexity to a neural network, holding the length of training constant while adding layers to a neural network will certainly result in you eventually observing a drop like this. To figure out whether that is the explanation for this particular observation, try increasing the number of iterations of training that you're using and see if it improves. Using a more intelligent stopping criterion is also a good option, but a simple increase in the cut-off will give you answers faster.
I have a simple task to classify people by their height and hair length to either MAN or WOMAN category using a neural network. Also teach it the pattern with some examples and then use it to classify on its own.
I have a basic understanding of neural networks but would really need some help here.
I know that each neuron divides the area to two subareas, basically that is why P = w0 + w1*x1 + w2*x2 + ... + wn*xn is being used here (weights are just moving the line if we consider geometric representation).
I do understand that each epoche should modify the weights to get closer to correct result, yet I have never program it and I am hopeless about how to start.
How should I proceed, meaning: How can I determine the threshold and how should I deal with the inputs?
It is not a homework rather than task for the ones who were interested. I am and I would like to understand it.
Looks like you are dealing with a simple Perceptron with a threshold activation function. Have a look at this question. Since you ARE using a bias neuron (w0), you would set the threshold to 0.
You then simply take the output of your network and compare it to 0, so you would e.g. output class 1 if x < 0 and class 2 if x > 0. You could model the case x=0 as "indistinct".
For learning the weights you need to apply the Delta Learning Rule which can be implemented very easily. But be careful: a perceptron with a simple threshold activation function can only be correct if your data are linearly separable. If you have more complex data you will need a Multilayer Perceptron and a nonlinear activation function like the Logistic Sigmoid Function.
Have a look at Geoffrey Hintons Coursera Course, Lecture 2 for details.
I've been working with machine learning lately (but I'm not an expert) but you should look at the Accord.NET framework. It contains all the common machine learning algorithme out of the box. So it's easy to take an existing samples and modify it instead of starting from scratch. Also, the developper of the framework is very helpful in the forum available on the same page.
With the available samples, you may also discover something better than neural network like the Kernel Support Vector Machine. If you stick to the neural network, have fun modifying all the different variables and by tryout and error you will understand how it work.
Have fun!
Since you said:
I know that each neuron divides the area to two subareas
&
weights are just moving the line if we consider geometric representation
I think you want to use perseptron or ADALINE neural networks. These neural networks can just classify linear separable patterns. since your input data is complicated, It's better to use a Multi layer Non-Linear Neural network. (my suggestion is a two layer neural network with tanh activation function) . For training these network you should use back propagation algorithm.
For answering to
how should I deal with the inputs?
I need to know more details about the inputs( Like: are they just height and hair length or there is more, what is their range and your resolution and etc.)
If you're dealing with just height and hair length I suggest that divide heights and length in some classes (for example 160cm-165cm, 165cm-170cm & etc.) and for each one of these classes set an On/Off input neuron. then put a hidden layer after all classes related to heights and another hidden layer after all classes related to hair length (tanh activation function). Number of neurons in these two hidden layer is determined based on number of training cases.
then take these two hidden layer output and send them to an aggregation layer with 1 output neuron.
I'm training a neural network with backpropagation algorithm and this is the chart of Overall Errors:
( I'm calculating Overall error by this formula : http://www.colinfahey.com/neural_network_with_back_propagation_learning/neural_network_with_back_propagation_learning_en.html Part 6.3 : Overall training error)
I have used Power Trendline and after calculations, I saw that if epoches = 13000 => overall error = 0.2
Isn't this too high?
Is this chart normal? Seems that the training process will take too long... Right? What should I do? Isn't there any faster way?
EDIT : My neural network has a hidden layer with 200 neurons. and my input and output layers have 10-12 neurons. My problem is clustering characters. (it clusters Persian characters into some clusters with supervised training)
So you are using a ANN with 200 input nodes with 10-12 hidden nodes in the hidden layer, what activation function are you using if any for your hidden layer and output layer?
Is this a standard back propagation training algorithm and what training function are you using?
Each type of training function will affect the speed of training and in some cases its ability to generalise, you don't want to train against your data such that your neural network is only good for your training data.
So ideally you want decent training data that could be a sub sample of your real data say 15%.
You could training your data using conjugate gradient based algorithm:
http://www.mathworks.co.uk/help/toolbox/nnet/ug/bss331l-1.html#bss331l-2
this will train your network quickly.
10-12 nodes may not be ideal for your data, you can try changing the number in blocks of 5 or add another layer, in general more layers will improve the ability of your network to classify your problem but will increase the computational complexity and hence slow down the training.
Presumably these 10-12 nodes are 'features' you are trying to classify?
If so, you may wish to normalise them, so rescale each to between 0 and 1 or -1 to 1 depending on your activation function (e.g. tan sigmoidal will produce values in range -1 to +1):
http://www.heatonresearch.com/node/706
You may also train a neural network to identify the ideal number of nodes you should have in your hidden layer.
I've read about neural network a little while ago and I understand how an ANN (especially a multilayer perceptron that learns via backpropagation) can learn to classify an event as true or false.
I think there are two ways :
1) You get one output neuron. It it's value is > 0.5 the events is likely true, if it's value is <=0.5 the event is likely to be false.
2) You get two output neurons, if the value of the first is > than the value of the second the event is likely true and vice versa.
In these case, the ANN tells you if an event is likely true or likely false. It does not tell how likely it is.
Is there a way to convert this value to some odds or to directly get odds out of the ANN. I'd like to get an output like "The event has a 84% probability to be true"
Once a NN has been trained, for eg. using backprogation as mentioned in the question (whereby the backprogation logic has "nudged" the weights in ways that minimize the error function) the weights associated with all individual inputs ("outside" inputs or intra-NN inputs) are fixed. The NN can then be used for classifying purposes.
Whereby the math (and the "options") during the learning phase can get a bit thick, it is relatively simple and straightfoward when operating as a classifier. The main algorithm is to compute an activation value for each neuron, as the sum of the input x weight for that neuron. This value is then fed to an activation function which purpose's is to normalize it and convert it to a boolean (in typical cases, as some networks do not have an all-or-nothing rule for some of their layers). The activation function can be more complex than you indicated, in particular it needn't be linear, but whatever its shape, typically sigmoid, it operate in the same fashion: figuring out where the activation fits on the curve, and if applicable, above or below a threshold. The basic algorithm then processes all neurons at a given layer before proceeding to the next.
With this in mind, the question of using the perceptron's ability to qualify its guess (or indeed guesses - plural) with a percentage value, finds an easy answer: you bet it can, its output(s) is real-valued (if anything in need of normalizing) before we convert it to a discrete value (a boolean or a category ID in the case of several categories), using the activation functions and the threshold/comparison methods described in the question.
So... How and Where do I get "my percentages"?... All depends on the NN implementation, and more importantly, the implementation dictates the type of normalization functions that can be used to bring activation values in the 0-1 range and in a fashion that the sum of all percentages "add up" to 1. In its simplest form, the activation function can be used to normalize the value and the weights of the input to the output layer can be used as factors to ensure the "add up" to 1 question (provided that these weights are indeed so normalized themselves).
Et voilà!
Claritication: (following Mathieu's note)
One doesn't need to change anything in the way the Neural Network itself works; the only thing needed is to somehow "hook into" the logic of output neurons to access the [real-valued] activation value they computed, or, possibly better, to access the real-valued output of the activation function, prior its boolean conversion (which is typically based on a threshold value or on some stochastic function).
In other words, the NN works as previously, neither its training nor recognition logic are altered, the inputs to the NN stay the same, as do the connections between various layers etc. We only get a copy of the real-valued activation of the neurons in the output layer, and we use this to compute a percentage. The actual formula for the percentage calculation depends on the nature of the activation value and its associated function (its scale, its range relative to other neurons' output etc.).
Here are a few simple cases (taken from the question's suggested output rules)
1) If there is a single output neuron: the ratio of the value provided by the activation function relative to the range of that function should do.
2) If there are two (or more output neurons), as with classifiers for example: If all output neurons have the same activation function, the percentage for a given neuron is that of its activation function value divided by the sum of all activation function values. If the activation functions vary, it becomes a case by case situation because the distinct activation functions may be indicative of a purposeful desire to give more weight to some of the neurons, and the percentage should respect this.
What you can do is to use a sigmoid transfer function on the output layer nodes (that accepts data ranges (-inf,inf) and outputs a value in [-1,1]).
Then by using the 1-of-n output encoding (one node for each class), you can map the range [-1,1] to [0,1] and use it as probability for each class value (note that this works naturally for more than just two classes).
The activation value of a single output neuron is a linearly weighted sum, and may be directly interpreted as an approximate probability if the network is trained to give outputs a range from 0 to 1. This would tend to be the case if the transfer function (or output function) in both the preceding stage and providing the final output is in the 0 to 1 range too (typically the sigmoidal logistic function). However, there is no guarantee that it will but repairs are possible. Moreover unless the sigmoids are logistic and the weights are constrained to be positive and sum to 1, it is unlikely. Generally a neural network will train in a more balanced way using the tanh sigmoid and weights and activations that range positive and negative (due to the symmetry of this model). Another factor is the prevalence of the class - if it is 50% then a 0.5 threshold is likely to be effective for logistic and a 0.0 threshold for tanh. The sigmoid is designed to push things towards the centre of the range (on backpropogation) and constrain it from going out of the range (in feedforward). The significance of the performance (with respect to the Bernoulli distribution) can also be interpreted as a probability that the neuron is making real predictions rather than guessing. Ideally the bias of the predictor to positives should match the prevalence of positives in the real world (which may vary at different times and places, e.g. bull vs bear markets, e.g. credit worthiness of people applying for loans vs people who fail to make loan payments) - calibrating to probabilities has the advantage that any desired bias can be set easily.
If you have two neurons for two classes, each can be interpreted independently as above, and the halved difference between them can also be. It is like flipping the negative class neuron and averaging. The differences can also give rise to a probability of significance estimate (using the T-test).
The Brier score and its Murphy decomposition give a more direct estimate of the probability that an average answer is correct, while Informedness gives the probability the classifier is making an informed decision rather than a guess, ROC AUC gives the probability a positive class will be ranked higher than a negative class (by a positive predictor), and Kappa will give a similar number that matches Informedness when prevalence = bias.
What you normally want is both a significance probability for the overall classifier (to ensure that you are playing on a real field, and not in an imaginary framework of guestimates) and a probability estimate for a specific example. There are various ways to calibrate, including doing a regression (linear or nonlinear) versus probability and using its inverse function to remap to a more accurate probability estimate. This can be seen by the Brier score improving, with the calibration component reducing towards 0, but the discrimination component remaining the same, as should ROC AUC and Informedness (Kappa is subject to bias and may worsen).
A simple non-linear way to calibrate to probabilities is to use the ROC curve - as the threshold changes for the output of a single neuron or the difference between two competing neurons, we plot the results true and false positive rates on a ROC curve (the false and true negative rates are naturally the complements, as what isn't really a positive is a negative). Then you scan the ROC curve (polyline) point by point (each time the gradient changes) sample by sample and the proportion of positive samples gives you a probability estimate for positives corresponding to the neural threshold that produced that point. Values between points on the curve can be linearly interpolated between those that are represented in the calibration set - and in fact any bad points in the ROC curve, represented by deconvexities (dents) can be smoothed over by the convex hull - probabilistically interpolating between the endpoints of the hull segment. Flach and Wu propose a technique that actually flips the segment, but this depends on information being used the wrong way round and although it could be used repeatedly for arbitrary improvement on the calibration set, it will be increasingly unlikely to generalize to a test situation.
(I came here looking for papers I'd seen ages ago on these ROC-based approaches - so this is from memory and without these lost references.)
I will be very prudent in interpreting the outputs of a neural networks (in fact any machine learning classifier) as a probability. The machine is trained to discriminate between classes, not to estimate the probability density. In fact, we don't have this information in the data, we have to infer it. For my experience I din't advice anyone to interpret directly the outputs as probabilities.
did you try prof. Hinton's suggestion of training the network with softmax activation function and cross entropy error?
as an example create a three layer network with the following:
linear neurons [ number of features ]
sigmoid neurons [ 3 x number of features ]
linear neurons [ number of classes ]
then train them with cross entropy error softmax transfer with your favourite optimizer stochastic descent/iprop plus/ grad descent. After training the output neurons should be normalized to sum of 1.
Please see http://en.wikipedia.org/wiki/Softmax_activation_function for details. Shark Machine Learning framework does provide Softmax feature through combining two models. And prof. Hinton an excellent online course # http://coursera.com regarding the details.
I can remember I saw an example of Neural network trained with back propagation to approximate the probability of an outcome in the book Introduction to the theory of neural computation (hertz krogh palmer). I think the key to the example was a special learning rule so that you didn't have to convert the output of a unit to probability, but instead you got automatically the probability as output.
If you have the opportunity, try to check that book.
(by the way, "boltzman machines", although less famous, are neural networks designed specifically to learn probability distributions, you may want to check them as well)
When using ANN for 2-class classification and logistic sigmoid activation function is used in the output layer, the output values could be interpreted as probabilities.
So if you choosing between 2 classes, you train using 1-of-C encoding, where 2 ANN outputs will have training values (1,0) and (0,1) for each of classes respectively.
To get probability of first class in percent, just multiply first ANN output to 100. To get probability of other class use the second output.
This could be generalized for multi-class classification using softmax activation function.
You can read more, including proofs of probabilistic interpretation here:
[1] Bishop, Christopher M. Neural networks for pattern recognition. Oxford university press, 1995.