So I read somewhere that we as humans don't understand what exactly happens in a neural network, we just know that a neuron does something using the biases and the inputs given to it and leads us to a specific output.
My question here is, do we understand (mathematically speaking) how X input leads the computer to give Y input? If we don't, then why don't we understand it?
Let X be an input Matrix and Y be the associated output vector (our target). Let theta be the parameter of our model, representing the weights and the bias of each neuron.
Mathematically, a neural network can be represented as a function f such as f(X, theta) = Y + epsilon. Epsilon is the error of the model. The goal is to find the value of theta that is minimizing epsilon. To do so, we just have to find the global minimum of the multivariate function epsilon(theta) = f(X, theta) - Y. This is an optimization problem that can be solved thanks to gradient descent. So yes, mathematically, we understand how X input leads the computer to give output Y: it is just a matter of finding the minimum of a function. Additionally, as the structure of a neural network is quite simple (linear layers + activation functions), we are able to calculate easily the derivatives of epsilon() and propagate them throw the network.
However, it's not because we can explain mathematically a neural network that we can interpret it. It's very difficult to know the specific role played by each neuron in the prediction. In contrary, decision trees are much more interpretable, as we know which feature was used to make a split at each node of the tree.
Related
I have a question on inverse prediction in Machine Learning/Data Science. Here I give a example to illustrate my question: I have 20 input features X = (x0, x1, ... x19) and 3 output variables Y = (y0, y1, y2). The number of training/test data usually small, such as <1000 items or even <100 in the training set.
In general, by using the machine learning toolbox (such as scikit learn), I can train the models (such as random forest, linear/polynomial regression and neural network) from X --> Y. But what I actually want to know is, for example, how should I set X, so that I can have y1 values in a specific range (for example y1 > 100).
Does anyone know how to solve this kind of "inverse prediction"? There are two ways in my mind:
Train the model in the normal way: X-->Y, then set a dense mesh in the high dimension X space. In this example, it is 20 dimensions. Then use all the point in this mesh as input data and throw them to the trained model. Select all the input points where the predicted y1 > 100. Finally, use some methods, such as clustering to look for some patterns in the selected data points.
Direct learn models from Y to X. Then, set a dense mesh in the high dimension Y space, where let y1 > 100. Then use the trained models to calculate the X data points.
The second method might be OK when the Y also have high dimensions. But usually, in my application, Y is very low-dimension and X is very high-dimension, which makes me think method 2 is not very practical.
Does anyone have any new thoughts? I think this should be somehow very common in industry and maybe some people meet similar situation before.
Thank you!
From what I understand of your needs, #1 is an excellent fit for this problem. I recommend that you use a simple binary classifier SVM to discriminate good/bad X vectors. SVM works well with high-dimensional spaces, and reading out the coefficients is easy in most SVM interfaces.
Similar note that may be useful:
In inverse/backward prediction, we can predict inversely with similar accuracy of direct/forward prediction of X--->Y and backward of Y--->X only just with solving the systems of equations X<---->Y assuming weights and intercepts. Also, usually, it is better for linear problems AX=B. Note that it is usually possible the Python code for inverse prediction has a considerable error while solving the system of equations (n*n) is better choice with suitable accuracy for that.
Regards
This is w.r.t a hybrid of ANN and logistic regression in a binary classification problem. For example in one of the papers I came across they state that "A hybrid model type is constructed by using the logistic regression model to calculate the probability of failure and then adding that value as an additional input variable into the ANN. This type of model is defined as a Plogit-ANN model".
So, for n input variables, I'm trying to understand how the additional input n+1 to a ANN is treated by the activation function (eg. a logit function) and in the summation of weights multiplied by inputs. Do we treat this probability variable n+1 as one of the standalone weights like a special type of b0 that we add in the sum of weights multiplied by inputs e.g. Summation for each Neuron = (Sum (Wi*Xi))+additional variable.
Thank you for your assistance.
According to the description provided the easiest way is to treat this as additional feature of your data. So you have a model that predicts something about your original dataset (probability of some additional thing), thus you get x -> f(x). You simply concatenate it to your feature vector so x' = [x1 x2 ... xk f(x)], and push it through the network.
However the described approach is quite naive, since you are doing these two things (training f and training neural net) completely independently), what might be more beneficial is to instead treat fitting f as an auxiliary loss and train your model jointly.
I am try to write a neural network class but I don't fully understand some aspects of it. I have two questions on the folling design.
Am I doing this correctly? Does the bias neuron need to connect to all of neurons (except those in the input layer) or just those in the hidden layer?
My second question is about calculation the output value. I'm using the equation below to calculate the output value of the neurons.
HiddenLayerFirstNeuron.Value =
(input1.Value * weight) + (input2.Value * weight) + (Bias.Value * weight)
After this equation, I'm calculating the activation and the result send the output. And output neurons doing same.
I'm not sure what I am do and I want to clear up problems.
Take a look at: http://deeplearning.net/tutorial/contents.html in theano. This explains everything you need to know for multi layer perceptron using theano (symbolic mathematic library).
The bias is usually connected to all hidden and output units.
Yes, you compute the input of activation function like summation of weight*output of previous layer neuron.
Good luck with development ;)
There should be a separate bias neuron for each hidden and the output layer. Think of the layers as a function applied to a first order polynomials such as f(m*x+b)=y where y is your output and f(x) your activation function. If you look at the the linear term you will recognize the b. This represents the bias and it behaves similar with neural network as with this simplification: It shifts the hyperplane up and down the in the space. Keep in mind that you will have one bias per layer connected to all neurons of that layer f((wi*xi+b)+...+(wn*xn+b)) with an initial value of 1. When it comes to gradient descent, you will have to train this neuron like a normal weight.
In my opinion should you apply the activation function to the output layer as well. This is how it's usually done with multilayer perceptrons. But it actually depends of what you want. If you, for example, use the logistic function as activation function and you want an output in the interval (0,1), then you have to apply your activation function to the output as well. Since a basic linear combination, as it is in your example, can theoretically go above the boundaries of the previously mentioned Intervall.
I understand neural networks with any number of hidden layers can approximate nonlinear functions, however, can it approximate:
f(x) = x^2
I can't think of how it could. It seems like a very obvious limitation of neural networks that can potentially limit what it can do. For example, because of this limitation, neural networks probably can't properly approximate many functions used in statistics like Exponential Moving Average, or even variance.
Speaking of moving average, can recurrent neural networks properly approximate that? I understand how a feedforward neural network or even a single linear neuron can output a moving average using the sliding window technique, but how would recurrent neural networks do it without X amount of hidden layers (X being the moving average size)?
Also, let us assume we don't know the original function f, which happens to get the average of the last 500 inputs, and then output a 1 if it's higher than 3, and 0 if it's not. But for a second, pretend we don't know that, it's a black box.
How would a recurrent neural network approximate that? We would first need to know how many timesteps it should have, which we don't. Perhaps a LSTM network could, but even then, what if it's not a simple moving average, it's an exponential moving average? I don't think even LSTM can do it.
Even worse still, what if f(x,x1) that we are trying to learn is simply
f(x,x1) = x * x1
That seems very simple and straightforward. Can a neural network learn it? I don't see how.
Am I missing something huge here or are machine learning algorithms extremely limited? Are there other learning techniques besides neural networks that can actually do any of this?
The key point to understand is compact:
Neural networks (as any other approximation structure like, polynomials, splines, or Radial Basis Functions) can approximate any continuous function only within a compact set.
In other words the theory states that, given:
A continuous function f(x),
A finite range for the input x, [a,b], and
A desired approximation accuracy ε>0,
then there exists a neural network that approximates f(x) with an approximation error less than ε, everywhere within [a,b].
Regarding your example of f(x) = x2, yes you can approximate it with a neural network within any finite range: [-1,1], [0, 1000], etc. To visualise this, imagine that you approximate f(x) within [-1,1] with a Step Function. Can you do it on paper? Note that if you make the steps narrow enough you can achieve any desired accuracy. The way neural networks approximate f(x) is not much different than this.
But again, there is no neural network (or any other approximation structure) with a finite number of parameters that can approximate f(x) = x2 for all x in [-∞, +∞].
The question is very legitimate and unfortunately many of the answers show how little practitioners seem to know about the theory of neural networks. The only rigorous theorem that exists about the ability of neural networks to approximate different kinds of functions is the Universal Approximation Theorem.
The UAT states that any continuous function on a compact domain can be approximated by a neural network with only one hidden layer provided the activation functions used are BOUNDED, continuous and monotonically increasing. Now, a finite sum of bounded functions is bounded by definition.
A polynomial is not bounded so the best we can do is provide a neural network approximation of that polynomial over a compact subset of R^n. Outside of this compact subset, the approximation will fail miserably as the polynomial will grow without bound. In other words, the neural network will work well on the training set but will not generalize!
The question is neither off-topic nor does it represent the OP's opinion.
I am not sure why there is such a visceral reaction, I think it is a legitimate question that is hard to find by googling it, even though I think it is widely appreciated and repeated outloud. I think in this case you are looking for the actually citations showing that a neural net can approximate any function. This recent paper explains it nicely, in my opinion. They also cite the original paper by Barron from 1993 that proved a less general result. The conclusion: a two-layer neural network can represent any bounded degree polynomial, under certain (seemingly non-restrictive) conditions.
Just in case the link does not work, it is called "Learning Polynomials with Neural Networks" by Andoni et al., 2014.
I understand neural networks with any number of hidden layers can approximate nonlinear functions, however, can it approximate:
f(x) = x^2
The only way I can make sense of that question is that you're talking about extrapolation. So e.g. given training samples in the range -1 < x < +1 can a neural network learn the right values for x > 100? Is that what you mean?
If you had prior knowledge, that the functions you're trying to approximate are likely to be low-order polynomials (or any other set of functions), then you could surely build a neural network that can represent these functions, and extrapolate x^2 everywhere.
If you don't have prior knowledge, things are a bit more difficult: There are infinitely many smooth functions that fit x^2 in the range -1..+1 perfectly, and there's no good reason why we would expect x^2 to give better predictions than any other function. In other words: If we had no prior knowledge about the function we're trying to learn, why would we want to learn x -> x^2? In the realm of artificial training sets, x^2 might be a likely function, but in the real world, it probably isn't.
To give an example: Let's say the temperature on Monday (t=0) is 0°, on Tuesday it's 1°, on Wednesday it's 4°. We have no reason to believe temperatures behave like low-order polynomials, so we wouldn't want to infer from that data that the temperature next Monday will probably be around 49°.
Also, let us assume we don't know the original function f, which happens to get the average of the last 500 inputs, and then output a 1 if it's higher than 3, and 0 if it's not. But for a second, pretend we don't know that, it's a black box.
How would a recurrent neural network approximate that?
I think that's two questions: First, can a neural network represent that function? I.e. is there a set of weights that would give exactly that behavior? It obviously depends on the network architecture, but I think we can come up with architectures that can represent (or at least closely approximate) this kind of function.
Question two: Can it learn this function, given enough training samples? Well, if your learning algorithm doesn't get stuck in a local minimum, sure: If you have enough training samples, any set of weights that doesn't approximate your function gives a training error greater that 0, while a set of weights that fit the function you're trying to learn has a training error=0. So if you find a global optimum, the network must fit the function.
A network can learn x|->x * x if it has a neuron that calculates x * x. Or more generally, a node that calculates x**p and learns p. These aren't commonly used, but the statement that "no neural network can learn..." is too strong.
A network with ReLUs and a linear output layer can learn x|->2*x, even on an unbounded range of x values. The error will be unbounded, but the proportional error will be bounded. Any function learnt by such a network is piecewise linear, and in particular asymptotically linear.
However, there is a risk with ReLUs: once a ReLU is off for all training examples it ceases learning. With a large domain, it will turn on for some possible test examples, and give an erroneous result. So ReLUs are only a good choice if test cases are likely to be within the convex hull of the training set. This is easier to guarantee if the dimensionality is low. One work around is to prefer LeakyReLU.
One other issue: how many neurons do you need to achieve the approximation you want? Each ReLU or LeakyReLU implements a single change of gradient. So the number needed depends on the maximum absolute value of the second differential of the objective function, divided by the maximum error to be tolerated.
There are theoretical limitations of Neural Networks. No neural network can ever learn the function f(x) = x*x
Nor can it learn an infinite number of other functions, unless you assume the impractical:
1- an infinite number of training examples
2- an infinite number of units
3- an infinite amount of time to converge
NNs are good in learning low-level pattern recognition problems (signals that in the end have some statistical pattern that can be represented by some "continuous" function!), but that's it!
No more!
Here's a hint:
Try to build a NN that takes n+1 data inputs (x0, x1, x2, ... xn) and it will return true (or 1) if (2 * x0) is in the rest of the sequence. And, good luck.
Infinite functions especially those that are recursive cannot be learned. They just are!
I have read through a lot of papers and understand the basic concept of a support vector machine at a very high level. You give it a training input vector which has a set of features and bases on how the "optimization function" evaluates this input vector lets call it x, (lets say we're talking about text classification), the text associated with the input vector x is classified into one of two pre-defined classes, this is only in the case of binary classification.
So my first question is through this procedure described above, all the papers say first that this training input vector x is mapped to a higher (maybe infinite) dimensional space. So what does this mapping achieve or why is this required? Lets say the input vector x has 5 features so who decides which "higher dimension" x is going to be mapped to?
Second question is about the following optimization equation:
min 1/2 wi(transpose)*wi + C Σi = 1..n ξi
so I understand that w has something to do with the margins of the hyperplane from the support vectors in the graph and I know that C is some sort of a penalty but I dont' know what it is a penalty for. And also what is ξi representing in this case.
A simple explanation of the second question would be much appreciated as I have not had much luck understanding it by reading technical papers.
When they talk about mapping to a higher-dimensional space, they mean that the kernel accomplishes the same thing as mapping the points to a higher-dimensional space and then taking dot products there. SVMs are fundamentally a linear classifier, but if you use kernels, they're linear in a space that's different from the original data space.
To be concrete, let's talk about the kernel
K(x, y) = (xy + 1)^2 = (xy)^2 + 2xy + 1,
where x and y are each real numbers (one-dimensional). Note that
(x^2, sqrt(2) x, 1) • (y^2, sqrt(2) y, 1) = x^2 y^2 + 2 x y + 1
has the same value. So K(x, y) = phi(x) • phi(y), where phi(a) = (a^2, sqrt(2), 1), and doing an SVM with this kernel (the inhomogeneous polynomial kernel of degree 2) is the same as if you first mapped your 1d points into this 3d space and then did a linear kernel.
The popular Gaussian RBF kernel function is equivalent to mapping your points into an infinite-dimensional Hilbert space.
You're the one who decides what feature space it's mapped into, when you pick a kernel. You don't necessarily need to think about the explicit mapping when you do that, though, and it's important to note that the data is never actually transformed into that high-dimensional space explicitly - then infinite-dimensional points would be hard to represent. :)
The ξ_i are the "slack variables". Without them, SVMs would never be able to account for training sets that aren't linearly separable -- which most real-world datasets aren't. The ξ in some sense are the amount you need to push data points on the wrong side of the margin over to the correct side. C is a parameter that determines how much it costs you to increase the ξ (that's why it's multiplied there).
1) The higher dimension space happens through the kernel mechanism. However, when evaluating the test sample, the higher dimension space need not be explicitly computed. (Clearly this must be the case because we cannot represent infinite dimensions on a computer.) For instance, radial basis function kernels imply infinite dimensional spaces, yet we don't need to map into this infinite dimension space explicitly. We only need to compute, K(x_sv,x_test), where x_sv is one of the support vectors and x_test is the test sample.
The specific higher dimensional space is chosen by the training procedure and parameters, which choose a set of support vectors and their corresponding weights.
2) C is the weight associated with the cost of not being able to classify the training set perfectly. The optimization equation says to trade-off between the two undesirable cases of non-perfect classification and low margin. The ξi variables represent by how much we're unable to classify instance i of the training set, i.e., the training error of instance i.
See Chris Burges' tutorial on SVM's for about the most intuitive explanation you're going to get of this stuff anywhere (IMO).