Closed. This question needs details or clarity. It is not currently accepting answers.
Want to improve this question? Add details and clarify the problem by editing this post.
Closed 5 years ago.
Improve this question
First, let me apologize for cramming three questions in that title. I'm not sure what better way is there.
I'll get right to it. I think I understand feedforward neural networks pretty well.
But LSTM really escapes me, and I feel maybe this is because I don't have a very good grasp of Recurrent neural networks in general. I have went through Hinton's and Andrew Ng's course on Coursera. A lot of it still doesn't make sense to me.
From what I understood, recurrent neural networks are different from feedforward neural networks in that past values influence the next prediction. Recurrent neural network are generally used for sequences.
The example I saw of recurrent neural network was binary addition.
010
+ 011
A recurrent neural network would take the right most 0 and 1 first, output a 1. Then take the 1,1 next, output a zero, and carry the 1. Take the next 0,0 and output a 1 because it carried the 1 from last calculation. Where does it store this 1? In feed forward networks the result is basically:
y = a(w*x + b)
where w = weights of connections to previous layer
and x = activation values of previous layer or inputs
How is a recurrent neural network calculated? I am probably wrong but from what I understood, recurrent neural networks are pretty much feedforward neural network with T hidden layers, T being number of timesteps. And each hidden layer takes the X input at timestep T and it's outputs are then added to the next respective hidden layer's inputs.
a(l) = a(w*x + b + pa)
where l = current timestep
and x = value at current timestep
and w = weights of connections to input layer
and pa = past activation values of hidden layer
such that neuron i in layer l uses the output value of neuron i in layer l-1
y = o(w*a(l-1) + b)
where w = weights of connections to last hidden layer
But even if I understood this correctly, I don't see the advantage of doing this over simply using past values as inputs to a normal feedforward network (sliding window or whatever it's called).
For example, what is the advantage of using a recurrent neural network for binary addition instead of than training a feedforward network with two output neurons. One for the binary result and the other for the carry? And then take the carry output and plug it back into the feedforward network.
However, I'm not sure how is this different than simply having past values as inputs in a feedforward model.
It seems to me that the more timesteps there are, recurrent neural networks are only a disadvantage over feedforward networks because of vanishing gradient. Which brings me to my second question, from what I understood, LSTM is a solution to the problem of vanishing gradient. But I have no actual grasp of how they work. Furthermore, are they simply better than recurrent neural networks, or are there sacrifices to using a LSTM?
What is a Recurrent neural network?
The basic idea is that recurrent networks have loops. These loops allow the network to use information from previous passes, which acts as memory. The length of this memory depends on a number of factors but it is important to note that it is not indefinite. You can think of the memory as degrading, with older information being less and less usable.
For example, let's say we just want the network to do one thing: Remember whether an input from earlier was 1, or 0. It's not difficult to imagine a network which just continually passes the 1 around in a loop. However every time you send in a 0, the output going into the loop gets a little lower (This is a simplification, but displays the idea). After some number of passes the loop input will be arbitrarily low, making the output of the network 0. As you are aware, the vanishing gradient problem is essentially the same, but in reverse.
Why not just use a window of time inputs?
You offer an alternative: A sliding window of past inputs being provided as current inputs. That's is not a bad idea, but consider this: While the RNN may have eroded over time, you will always lose the entirety of your time information after you window ends. And while you would remove the vanishing gradient problem, you would have to increase the number of weights of your network by several times. Having to train all those additional weights will hurt you just as badly as (if not worse than) vanishing gradient.
What is an LSTM network?
You can think of LSTM as a special type of RNN. The difference is that LSTM is able to actively maintain self connecting loops without them degrading. This is accomplished through a somewhat fancy activation, involving an additional "memory" output for the self looping connection. The network must then be trained to select what data gets put onto this bus. By training the network to explicit select what to remember, we don't have to worry about new inputs destroying important information, and the vanishing gradient doesn't affect the information we decided to keep.
There are two main drawbacks:
It is more expensive to calculate the network output and apply back propagation. You simply have more math to do because of the complex activation. However this is not as important as the second point.
The explicit memory adds several more weights to each node, all of which must be trained. This increases the dimensionality of the problem, and potentially makes it harder to find an optimal solution.
Is it always better?
Which structure is better depends on a number of factors, like the number of nodes you need for you problem, the amount of available data, and how far back you want your network's memory to reach. However if you only want the theoretical answer, I would say that given infinite data and computing speed, an LSTM is the better choice, however one should not take this as practical advice.
A feed forward neural network has connections from layer n to layer n+1.
A recurrent neural network allows connections from layer n to layer n as well.
These loops allow the network to perform computations on data from previous cycles, which creates a network memory. The length of this memory depends on a number of factors and is an area of active research, but could be anywhere from tens to hundreds of time steps.
To make it a bit more clear, the carried 1 in your example is stored in the same way as the inputs: in a pattern of activation of a neural layer. It's just the recurrent (same layer) connections that allow the 1 to persist through time.
Obviously it would be infeasible to replicate every input stream for more than a few past time steps, and choosing which historical streams are important would be very difficult (and lead to reduced flexibility).
LSTM is a very different model which I'm only familiar with by comparison to the PBWM model, but in that review LSTM was able to actively maintain neural representations indefinitely, so I believe it is more intended for explicit storage. RNNs are more suited to non-linear time series learning, not storage. I don't know if there are drawbacks to using LSTM rather RNNs.
Both RNN and LSTM can be sequence learners. RNN suffers from vanishing gradient point problem. This problem causes the RNN to have trouble in remembering values of past inputs after more than 10 timesteps approx. (RNN can remember previously seen inputs for a few time steps only)
LSTM is designed to solve the vanishing gradient point problem in RNN. LSTM has the capability of bridging long time lags between inputs. In other words, it is able to remember inputs from up to 1000 time steps in the past (some papers even made claims it can go more than this). This capability makes LSTM an advantage for learning long sequences with long time lags. Refer to Alex Graves Ph.D. thesis Supervised Sequence Labelling
with Recurrent Neural Networks for some details. If you are new to LSTM, I recommend Colah's blog for super simple and easy explanation.
However, recent advances in RNN also claim that with careful initialization, RNN can also learn long sequences comparable to the performance of LSTM. A Simple Way to Initialize Recurrent Networks of Rectified Linear Units.
Related
I am currently looking through Michael Nielsen's ebook Neural Networks and Deep Learning and have run the code found at the end of chapter 1 which trains a neural network to recognize hand-written digits (with a slight modification to make the backpropagation algorithm over a mini-batch matrix-based).
However, having run this code and achieving a classification accuracy of just under 94%, I decided to remove the use of biases from the network. After re-training the modified network, I found no difference in classification accuracy!
NB: The output layer of this network contains ten neurons; if the ith of these neurons has the highest activation then the input is classified as being the digit i.
This got me wondering why it is necessary to use biases in a neural network, rather than just weights, and what differentiates between a task where biases will improve the performance of a network and a task where they will not?
My code can be found here: https://github.com/pipthagoras/neural-network-1
Biases are used to account for the fact that your underlying data might not be centered. It is clearer to see in the case of a linear regression.
If you do a regression without an intercept (or bias), you are forcing the underlying model to pass through the origin, which will result in a poor model if the underlying data is not centered (for example if the true generating process is Y=3000). If, on the other hand, your data is centered or close to centered, then eliminating bias is good, since you won't introduce a term that is, in fact, independent to your predictive variable (it's like selecting a simpler model, which will tend to generalize better PROVIDED that it actually reflects the underlying data).
I am using a Bike Sharing dataset to predict the number of rentals in a day, given the input. I will use 2011 data to train and 2012 data to validate. I successfully built a linear regression model, but now I am trying to figure out how to predict time series by using Recurrent Neural Networks.
Data set has 10 attributes (such as month, working day or not, temperature, humidity, windspeed), all numerical, though an attribute is day (Sunday: 0, Monday:1 etc.).
I assume that one day can and probably will depend on previous days (and I will not need all 10 attributes), so I thought about using RNN. I don't know much, but I read some stuff and also this. I think about a structure like this.
I will have 10 input neurons, a hidden layer and 1 output neuron. I don't know how to decide on how many neurons the hidden layer will have.
I guess that I need a matrix to connect input layer to hidden layer, a matrix to connect hidden layer to output layer, and a matrix to connect hidden layers in neighbouring time-steps, t-1 to t, t to t+1. That's total of 3 matrices.
In one tutorial, activation function was sigmoid, although I'm not sure exactly, if I use sigmoid function, I will only get output between 0 and 1. What should I use as activation function? My plan is to repeat this for n times:
For each training data:
Forward propagate
Propagate the input to hidden layer, add it to propagation of previous hidden layer to current hidden layer. And pass this to activation function.
Propagate the hidden layer to output.
Find error and its derivative, store it in a list
Back propagate
Find current layers and errors from list
Find current hidden layer error
Store weight updates
Update weights (matrices) by multiplying them by learning rate.
Is this the correct way to do it? I want real numerical values as output, instead of a number between 0-1.
It seems to be the correct way to do it, if you are just wanting to learn the basics. If you want to build a neural network for practical use, this is a very poor approach and as Marcin's comment says, almost everyone who constructs neural nets for practical use do so by using packages which have an ready simulation of neural network available. Let me answer your questions one by one...
I don't know how to decide on how many neurons the hidden layer will have.
There is no golden rule to choose the right architecture for your neural network. There are many empirical rules people have established out of experience, and the right number of neurons are decided by trying out various combinations and comparing the output. A good starting point would be (3/2 times your input plus output neurons, i.e. (10+1)*(3/2)... so you could start with a 15/16 neurons in hidden layer, and then go on reducing the number based on your output.)
What should I use as activation function?
Again, there is no 'right' function. It totally depends on what suits your data. Additionally, there are many types of sigmoid functions like hyperbolic tangent, logistic, RBF, etc. A good starting point would be logistic function, but again you will only find the right function through trial and error.
Is this the correct way to do it? I want real numerical values as output, instead of a number between 0-1.
All activation functions(including the one assigned to output neuron) will give you an output of 0 to 1, and you will have to use multiplier to convert it to real values, or have some kind of encoding with multiple output neurons. Coding this manually will be complicated.
Another aspect to consider would be your training iterations. Doing it 'n' times doesn't help. You need to find the optimal training iterations with trial and error as well to avoid both under-fitting and over-fitting.
The correct way to do it would be to use packages in Python or R, which will allow you to train neural nets with large amount of customization quickly, where you can train and test multiple nets with different activation functions (and even different training algorithms) and network architecture without too much hassle. With some amount of trial and error, you will eventually find the net that gives you desirable output.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed last year.
Improve this question
A simple example: Given an input sequence, I want the neural network to output the median of the sequence. The problem is, if a neural network learnt to compute the median of n inputs, how can it compute the median of even more inputs? I know that recurrent neural networks can learn functions like max and parity over a sequence, but computing these functions only requires constant memory. What if the memory requirement grows with the input size like computing the median?
This is a follow up question on How are neural networks used when the number of inputs could be variable?.
One idea I had is the following: treating each weight as a function of the number of inputs instead of a fixed value. So a weight may have many parameters that define a function, and we train these parameters. For example, if we want the neural network to compute the average of n inputs, we would like each weight function behaves like 1/n. Again, average per se can be computed using recurrent neural networks or hidden markov model, but I was hoping this kind of approaches can be generalized to solve certain problems where memory requirement grows.
If a neural network learnt to compute the median of n inputs, how can it compute the median of even more inputs?
First of all, you should understand the use of a neural network. We, generally use the neural network in problems where a mathematical solution is not possible. In this problem, use of NN is not significant/ unadvisable.
There are other problems of such nature, like forecasting, in which continuous data arrives over time.
One solution to such problem can be Hidden Markov Model (HMM). But again, such models depends on the correlation between input over a period of time. So This model is not efficient for problems where the input is completely random.
So, If input is completely random and memory requirement grows
There is nothing much you can do about it, one possible solution could be growing your memory size.
Just remember one thing, NN and similar models of machine learning aims to extract meaningful information from the data. if data is just some random values then all models will generate some random output.
One more idea: some data transformation. Let have N big enough that always bigger than n. We make a net with 2*N inputs. First N inputs are for data. If n less then N, then rest inputs set to 0. Last N inputs are intended for specifying which numbers are useful. Thus 1 is data, 0 is not data. As follows in Matlab notation: if v is an input, and it is a vector of length 2*N, then we put into v(1:n) our original data. After that, we put to v(n+1:N) zeros. Then put to v(N+1:N+n) ones, and then put V(N+n+1:2*N) zeros. It is just an idea, which I have not checked. If you are interested in the application of neural networks, take a look at the example of how we have chosen an appropriate machine learning algorithm to classify EEG signals for BCI.
I Wrote a simple recurrent neural network (7 neurons, each one is initially connected to all the neurons) and trained it using a genetic algorithm to learn "complicated", non-linear functions like 1/(1+x^2). As the training set, I used 20 values within the range [-5,5] (I tried to use more than 20 but the results were not changed dramatically).
The network can learn this range pretty well, and when given examples of other points within this range, it can predict the value of the function. However, it can not extrapolate correctly and predicting the values of the function outside the range [-5,5]. What are the reasons for that and what can I do to improve its extrapolation abilities?
Thanks!
Neural networks are not extrapolation methods (no matter - recurrent or not), this is completely out of their capabilities. They are used to fit a function on the provided data, they are completely free to build model outside the subspace populated with training points. So in non very strict sense one should think about them as an interpolation method.
To make things clear, neural network should be capable of generalizing the function inside subspace spanned by the training samples, but not outside of it
Neural network is trained only in the sense of consistency with training samples, while extrapolation is something completely different. Simple example from "H.Lohninger: Teach/Me Data Analysis, Springer-Verlag, Berlin-New York-Tokyo, 1999. ISBN 3-540-14743-8" shows how NN behave in this context
All of these networks are consistent with training data, but can do anything outside of this subspace.
You should rather reconsider your problem's formulation, and if it can be expressed as a regression or classification problem then you can use NN, otherwise you should think about some completely different approach.
The only thing, which can be done to somehow "correct" what is happening outside the training set is to:
add artificial training points in the desired subspace (but this simply grows the training set, and again - outside of this new set, network's behavious is "random")
add strong regularization, which will force network to create very simple model, but model's complexity will not guarantee any extrapolation strength, as two model's of exactly the same complexity can have for example completely different limits in -/+ infinity.
Combining above two steps can help building model which to some extent "extrapolates", but this, as stated before, is not a purpose of a neural network.
As far as I know this is only possible with networks which do have the echo property. See Echo State Networks on scholarpedia.org.
These networks are designed for arbitrary signal learning and are capable to remember their behavior.
You can also take a look at this tutorial.
The nature of your post(s) suggests that what you're referring to as "extrapolation" would be more accurately defined as "sequence recognition and reproduction." Training networks to recognize a data sequence with or without time-series (dt) is pretty much the purpose of Recurrent Neural Network (RNN).
The training function shown in your post has output limits governed by 0 and 1 (or -1, since x is effectively abs(x) in the context of that function). So, first things first, be certain your input layer can easily distinguish between negative and positive inputs (if it must).
Next, the number of neurons is not nearly as important as how they're layered and interconnected. How many of the 7 were used for the sequence inputs? What type of network was used and how was it configured? Network feedback will reveal the ratios, proportions, relationships, etc. and aid in the adjustment of network weight adjustments to match the sequence. Feedback can also take the form of a forward-feed depending on the type of network used to create the RNN.
Producing an 'observable' network for the exponential-decay function: 1/(1+x^2), should be a decent exercise to cut your teeth on RNNs. 'Observable', meaning the network is capable of producing results for any input value(s) even though its training data is (far) smaller than all possible inputs. I can only assume that this was your actual objective as opposed to "extrapolation."
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
This question does not appear to be about programming within the scope defined in the help center.
Closed 2 years ago.
Improve this question
Why do we have to normalize the input for a neural network?
I understand that sometimes, when for example the input values are non-numerical a certain transformation must be performed, but when we have a numerical input? Why the numbers must be in a certain interval?
What will happen if the data is not normalized?
It's explained well here.
If the input variables are combined linearly, as in an MLP [multilayer perceptron], then it is
rarely strictly necessary to standardize the inputs, at least in theory. The
reason is that any rescaling of an input vector can be effectively undone by
changing the corresponding weights and biases, leaving you with the exact
same outputs as you had before. However, there are a variety of practical
reasons why standardizing the inputs can make training faster and reduce the
chances of getting stuck in local optima. Also, weight decay and Bayesian
estimation can be done more conveniently with standardized inputs.
In neural networks, it is good idea not just to normalize data but also to scale them. This is intended for faster approaching to global minima at error surface. See the following pictures:
Pictures are taken from the coursera course about neural networks. Author of the course is Geoffrey Hinton.
Some inputs to NN might not have a 'naturally defined' range of values. For example, the average value might be slowly, but continuously increasing over time (for example a number of records in the database).
In such case feeding this raw value into your network will not work very well. You will teach your network on values from lower part of range, while the actual inputs will be from the higher part of this range (and quite possibly above range, that the network has learned to work with).
You should normalize this value. You could for example tell the network by how much the value has changed since the previous input. This increment usually can be defined with high probability in a specific range, which makes it a good input for network.
There are 2 Reasons why we have to Normalize Input Features before Feeding them to Neural Network:
Reason 1: If a Feature in the Dataset is big in scale compared to others then this big scaled feature becomes dominating and as a result of that, Predictions of the Neural Network will not be Accurate.
Example: In case of Employee Data, if we consider Age and Salary, Age will be a Two Digit Number while Salary can be 7 or 8 Digit (1 Million, etc..). In that Case, Salary will Dominate the Prediction of the Neural Network. But if we Normalize those Features, Values of both the Features will lie in the Range from (0 to 1).
Reason 2: Front Propagation of Neural Networks involves the Dot Product of Weights with Input Features. So, if the Values are very high (for Image and Non-Image Data), Calculation of Output takes a lot of Computation Time as well as Memory. Same is the case during Back Propagation. Consequently, Model Converges slowly, if the Inputs are not Normalized.
Example: If we perform Image Classification, Size of Image will be very huge, as the Value of each Pixel ranges from 0 to 255. Normalization in this case is very important.
Mentioned below are the instances where Normalization is very important:
K-Means
K-Nearest-Neighbours
Principal Component Analysis (PCA)
Gradient Descent
When you use unnormalized input features, the loss function is likely to have very elongated valleys. When optimizing with gradient descent, this becomes an issue because the gradient will be steep with respect some of the parameters. That leads to large oscillations in the search space, as you are bouncing between steep slopes. To compensate, you have to stabilize optimization with small learning rates.
Consider features x1 and x2, where range from 0 to 1 and 0 to 1 million, respectively. It turns out the ratios for the corresponding parameters (say, w1 and w2) will also be large.
Normalizing tends to make the loss function more symmetrical/spherical. These are easier to optimize because the gradients tend to point towards the global minimum and you can take larger steps.
Looking at the neural network from the outside, it is just a function that takes some arguments and produces a result. As with all functions, it has a domain (i.e. a set of legal arguments). You have to normalize the values that you want to pass to the neural net in order to make sure it is in the domain. As with all functions, if the arguments are not in the domain, the result is not guaranteed to be appropriate.
The exact behavior of the neural net on arguments outside of the domain depends on the implementation of the neural net. But overall, the result is useless if the arguments are not within the domain.
I believe the answer is dependent on the scenario.
Consider NN (neural network) as an operator F, so that F(input) = output. In the case where this relation is linear so that F(A * input) = A * output, then you might choose to either leave the input/output unnormalised in their raw forms, or normalise both to eliminate A. Obviously this linearity assumption is violated in classification tasks, or nearly any task that outputs a probability, where F(A * input) = 1 * output
In practice, normalisation allows non-fittable networks to be fittable, which is crucial to experimenters/programmers. Nevertheless, the precise impact of normalisation will depend not only on the network architecture/algorithm, but also on the statistical prior for the input and output.
What's more, NN is often implemented to solve very difficult problems in a black-box fashion, which means the underlying problem may have a very poor statistical formulation, making it hard to evaluate the impact of normalisation, causing the technical advantage (becoming fittable) to dominate over its impact on the statistics.
In statistical sense, normalisation removes variation that is believed to be non-causal in predicting the output, so as to prevent NN from learning this variation as a predictor (NN does not see this variation, hence cannot use it).
The reason normalization is needed is because if you look at how an adaptive step proceeds in one place in the domain of the function, and you just simply transport the problem to the equivalent of the same step translated by some large value in some direction in the domain, then you get different results. It boils down to the question of adapting a linear piece to a data point. How much should the piece move without turning and how much should it turn in response to that one training point? It makes no sense to have a changed adaptation procedure in different parts of the domain! So normalization is required to reduce the difference in the training result. I haven't got this written up, but you can just look at the math for a simple linear function and how it is trained by one training point in two different places. This problem may have been corrected in some places, but I am not familiar with them. In ALNs, the problem has been corrected and I can send you a paper if you write to wwarmstrong AT shaw.ca
On a high level, if you observe as to where normalization/standardization is mostly used, you will notice that, anytime there is a use of magnitude difference in model building process, it becomes necessary to standardize the inputs so as to ensure that important inputs with small magnitude don't loose their significance midway the model building process.
example:
√(3-1)^2+(1000-900)^2 ≈ √(1000-900)^2
Here, (3-1) contributes hardly a thing to the result and hence the input corresponding to these values is considered futile by the model.
Consider the following:
Clustering uses euclidean or, other distance measures.
NNs use optimization algorithm to minimise cost function(ex. - MSE).
Both distance measure(Clustering) and cost function(NNs) use magnitude difference in some way and hence standardization ensures that magnitude difference doesn't command over important input parameters and the algorithm works as expected.
Hidden layers are used in accordance with the complexity of our data. If we have input data which is linearly separable then we need not to use hidden layer e.g. OR gate but if we have a non linearly seperable data then we need to use hidden layer for example ExOR logical gate.
Number of nodes taken at any layer depends upon the degree of cross validation of our output.