After profiling my Neural Nets' code I've realized that the method, which computes the weight changes for each arc in the network (-rate*gradient + momentum*previous_delta - decay*rate*weight), already given the gradient, is the bottleneck (55% inclusive samples).
Is there any trick to compute these values in a efficient manner?
This is normal behaviour. I am assuming that you are using an iterative process to solve the weights at each evolution step (such as backpropagation?). If the number of neurons is large and the training (back-testing) algorithm is short, then it is normal that weight mutation such as this will consume a larger fraction of compute time during training of the neural network.
Did you get this result using a simple XOR problem or similar? If so, you will probably find that if you start to solve more complex problems (such as pattern detection in multidimensional arrays, image processing, etc.) that those functions will begin to consume an insignificant fraction of compute time.
If you are profiling, I would suggest you profile with a problem that is closer to the purpose for which the neural network is designed (I am guessing you didn't design it to solve XOR or play tic tac toe) and you will probably find that optimising code such as -rate*gradient + momentum*previous_delta - decay*rate*weight is more or less a waste of time, at least this is my experience.
If you do find that this code is compute-intensive in real-world applications then I would suggest trying to reduce the number of times this line of code is executed via structural changes. Neural network optimization is a rich field and I can't possibly give you useful advise from such a broad question, but I will say that if your program is unusually slow, you're unlikely to see significant improvements by tinkering at such low-level code. I will however suggest the following from my own experience:
Consider parallelisation. Many search algorithms such as those implemented in back-propagation techniques are amenable to parallel attempts to improve convergence. As weight-adjustments are identical in terms of computation demand for a given network, think static loops in Open MP.
Modify the convergence criterion (the critical convergence rate before you stop adjustments of weights) to perform less of these calculations
Consider an alternative to deterministic solutions such as back-propagations, which are slightly more prone to local optimisation anyway. Consider gaussian mutation (All things being equal gaussian mutation will 1) reduce time spent on mutation relative to backtesting 2) increase convergence time and 3) be less prone to getting caught in local minima of the error search space)
Please note that this is a non-technical answer to what I have interpreted as a non-technical question.
Related
I've seen some comments in online articles/tutorials or Stack Overflow questions which suggest that increasing number of epochs can result in overfitting. But my intuition tells me that there should be no direct relationship at all between number of epochs and overfitting. So I'm looking for answer which explains if I'm right or wrong (or whatever's in between).
Here's my reasoning though. To overfit, you need to have enough free parameters (I think this is called "capacity" in neural networks) in your model to generate a function which can replicate the sample data points. If you don't have enough free parameters, you'll never overfit. You might just underfit.
So really, if you don't have too many free parameters, you could run infinite epochs and never overfit. If you have too many free parameters, then yes, the more epochs you have the more likely it is that you get to a place where you're overfitting. But that's just because running more epochs revealed the root cause: too many free parameters. The real loss function doesn't care about how many epochs you run. It existed the moment you defined your model structure, before you ever even tried to do gradient descent on it.
In fact, I'd venture as far as to say: assuming you have the computational resources and time, you should always aim to run as many epochs as possible, because that will tell you whether your model is prone to overfitting. Your best model will be the one that provides great training and validation accuracy, no matter how many epochs you run it for.
EDIT
While reading more into this, I realise I forgot to take into account that you can arbitrarily vary the sample size as well. Given a fixed model, a smaller sample size is more prone to being overfit. And then that kind of makes me doubt my intuition above. Still happy to get an answer though!
Your intuition to me seems completely correct.
But here is the caveat. The whole purpose of deep models is that they are "deep" (duh!!). So what happens is that your feature space gets exponentially larger as you grow your network.
Here is an example to compare a deep model with a simpler mode:
Assume you have a 10-variable data set. With a crazy amount of feature engineering, you might be able to extract 50 features out of it. Then if you run a traditional model (let's say a logistic regression), you will have 50 parameters (capacity in your word, or degree of freedom) to train.
But, if you use a very simple deep model with Layer 1: 10 unit, layer2: 10 units, layer3: 5 units, layer4: 2 units, you will end up with (10*10 + 10*10 + 5*2 = 210) parameters to train.
Therefore, usually when we train a neural net for a long time, we end of with a memorized version of our data set(this gets worse if our data set is small and easy to be memorized).
But as you also mentioned, there is no intrinsic reason why higher number of epochs result in overfitting. Early stopping is usually a very good way for avoiding this. Just set patience equal to 5-10 epochs.
If the amount of trainable parameters is small with respect to the size of your training set (and your training set is reasonably diverse) then running over the same data multiple times will not be that significant, since you will be learning some features about your problem, rather than just memorizing the training data set. The problem arises when the amount of parameters is comparable to your training data set size (or bigger), it is basically the same problem as with any machine learning technique that uses too many features. This is quite common if you use large layers with dense connections. To combat this overfitting problem there are lots of regularization techniques (dropout, L1 regularizer, constraining certain connections to be 0 or equal such as in CNN).
The problem is that might still be left with too many trainable parameters. A simple way to regularize even further is to have a small learning rate (i.e. don't learn too much from this particular example lest you memorize it) combined with monitoring the epochs (if there is a large gap increase between validation/training accuracy, you are starting to overfit your model). You can then use the gap info to stop your training. This is a version of what is known as early stopping (stop before you reach the minimum in your loss function).
I am currently developing a deep reinforcement learning network however, I have a small doubt about the number of q-values I will have at the output of the NN. I will have a total of 150 q-values, which personally seems excessive to me. I have read on several papers and books that this could be a problem. I know that it will depend from the kind of NN I will build, but do you guys think that the number of q-values is too high? should I reduce it?
There is no general principle what is "too much". Everything depends solely on the problem and throughput one can get in learning. In particular number of actions does not have to matter as long as internal parametrisation of Q(a, s) is efficient. To give some example lets assume that the neural network is actually of form NN(a, s) = Q(a, s), in other words it accepts action as an input, together with the state, and outputs the Q value. If such an architecture can be trained in a problem considered, than it might be able to scale to big action spaces; on the other hand if the neural net basically has independent output per action, something of form NN(s)[a] = Q(a, s) then many actions can lead to relatively sparse learning signal for the model and thus lead to slow convergence.
Since you are asking about reducing action space it sounds like the true problem has complex control (maybe it is a continuous control domain?) and you are looking for some discretization to make it simpler to learn. If this is the case you will have to follow the typical approach of trial and error - try with simple action space, observe the dynamics, and if the results are not satisfactory - increase the complexity of the problem. This allows making iterative improvements, as opposed to going in the opposite direction - starting with too complex setting to get any results and than having to reduce it without knowing what are the "reasonable values".
I have gone through neural networks and have understood the derivation for back propagation almost perfectly(finally!). However, I had a small doubt.
We are updating all the weights simultaneously, so what is the guarantee that they lead to a smaller cost. If the weights are updated one by one, it would definitely lead to a lesser cost and it would be similar to linear regression. But if you update all the weights simultaneously, might we not cross the minima?
Also, do we update the biases like we update the weights after each forward propagation and back propagation of each test case?
Lastly, I have started reading on RNN's. What are some good resources to understand BPTT in RNN's?
Yes, updating only one weight at the time could result in decreasing error value every time but it's usually infeasible to do such updates in practical solutions using NN. Most of today's architectures usually have ~ 10^6 parameters so one epoch for every parameter could last enormously long. Moreover - because of nature of backpropagation - you usually have to compute loads of different derivatives in order to compute derivative with respect to a parameter given - so you will waste a lot of computations when using such approach.
But the phenomenon which you mention has been noticed a long time ago and there are some ways in dealing with it. There are two most common issues connected with it:
Covariance shift: it's when error and weight updates of a layer given strongly depends on output from previous layer, so when you update it - the results in the next layer might be different. The most common way to deal with this problem right now is Batch normalization.
Nolinear function vs Linear Differentation: it's quite uncommon when you think about BP but derivative is a linear operator which might generate a lot of problems in gradient descent. The most countintuitive example is the fact that if you multiply your input by a constant then every derivative will also be multiplied by the same number. This may lead to a lot of problems but most of recent methods of learning do a great job in dealing with it.
About BPTT I stronly recomend you Geoffrey Hinton course about ANN and especially this video.
i am experimenting with neural networks. I have a network with 8 input neurons, 5 hidden and 2 output. When i let the network learn with backpropagation, sometimes, it produces worse result between single iterations of training. What can be the cause? It should not be implementation error, because i even tried using implementation from Introduction to Neural Networks for Java and it does exactly the same.
Nothing is wrong. Back propagation is just a gradient optimization, and gradient methods do not have a guarantee of making error smaller in each iteration (you do have a guarantee that there exists a very small step size/learning rate which has such property, but in practise no way of finding it); furthermore you are probably updating weights after each sample making your training stochastic, which is even more "unstable" in this matter (as you do not really calculate the true gradient). However, if due to this, your method is not converging - think about proper scaling of your data as well as reducing the learning rate and probably adding the momentum term. These are just gradient-based optimization-related issues, not BP as such.
Is a genetic algorithm the most efficient way to optimize the number of hidden nodes and the amount of training done on an artificial neural network?
I am coding neural networks using the NNToolbox in Matlab. I am open to any other suggestions of optimization techniques, but I'm most familiar with GA's.
Actually, there are multiple things that you can optimize using GA regarding NN.
You can optimize the structure (number of nodes, layers, activation function etc.).
You can also train using GA, that means setting the weights.
Genetic algorithms will never be the most efficient, but they usually used when you have little clue as to what numbers to use.
For training, you can use other algorithms including backpropagation, nelder-mead etc..
You said you wanted to optimize number hidden nodes, for this, genetic algorithm may be sufficient, although far from "optimal". The space you are searching is probably too small to use genetic algorithms, but they can still work and afaik, they are already implemented in matlab, so no biggie.
What do you mean by optimizing amount of training done? If you mean number of epochs, then that's fine, just remember that training is somehow dependent on starting weights and they are usually random, so the fitness function used for GA won't really be a function.
A good example of neural networks and genetic programming is the NEAT architecture (Neuro-Evolution of Augmenting Topologies). This is a genetic algorithm that finds an optimal topology. It's also known to be good at keeping the number of hidden nodes down.
They also made a game using this called Nero. Quite unique and very amazing tangible results.
Dr. Stanley's homepage:
http://www.cs.ucf.edu/~kstanley/
Here you'll find just about everything NEAT related as he is the one who invented it.
Genetic algorithms can be usefully applied to optimising neural networks, but you have to think a little about what you want to do.
Most "classic" NN training algorithms, such as Back-Propagation, only optimise the weights of the neurons. Genetic algorithms can optimise the weights, but this will typically be inefficient. However, as you were asking, they can optimise the topology of the network and also the parameters for your training algorithm. You'll have to be especially wary of creating networks that are "over-trained" though.
One further technique with a modified genetic algorithms can be useful for overcoming a problem with Back-Propagation. Back-Propagation usually finds local minima, but it finds them accurately and rapidly. Combining a Genetic Algorithm with Back-Propagation, e.g., in a Lamarckian GA, gives the advantages of both. This technique is briefly described during the GAUL tutorial
It is sometimes useful to use a genetic algorithm to train a neural network when your objective function isn't continuous.
I'm not sure whether you should use a genetic algorithm for this.
I suppose the initial solution population for your genetic algorithm would consist of training sets for your neural network (given a specific training method). Usually the initial solution population consists of random solutions to your problem. However, random training sets would not really train your neural network.
The evaluation algorithm for your genetic algorithm would be a weighed average of the amount of training needed, the quality of the neural network in solving a specific problem and the numer of hidden nodes.
So, if you run this, you would get the training set that delivered the best result in terms of neural network quality (= training time, number hidden nodes, problem solving capabilities of the network).
Or are you considering an entirely different approach?
I'm not entirely sure what kind of problem you're working with, but GA sounds like a little bit of overkill here. Depending on the range of parameters you're working with, an exhaustive (or otherwise unintelligent) search may work. Try plotting your NN's performance with respect to number of hidden nodes for a first few values, starting small and jumping by larger and larger increments. In my experience, many NNs plateau in performance surprisingly early; you may be able to get a good picture of what range of hidden node numbers makes the most sense.
The same is often true for NNs' training iterations. More training helps networks up to a point, but soon ceases to have much effect.
In the majority of cases, these NN parameters don't affect performance in a very complex way. Generally, increasing them increases performance for a while but then diminishing returns kick in. GA is not really necessary to find a good value on this kind of simple curve; if the number of hidden nodes (or training iterations) really does cause the performance to fluctuate in a complicated way, then metaheuristics like GA may be apt. But give the brute-force approach a try before taking that route.
I would tend to say that genetic algorithms is a good idea since you can start with a minimal solution and grow the number of neurons. It is very likely that the "quality function" for which you want to find the optimal point is smooth and has only few bumps.
If you have to find this optimal NN frequently I would recommend using optimization algorithms and in your case quasi newton as described in numerical recipes which is optimal for problems where the function is expensive to evaluate.