I've been using tensorflow on and off for various things that I guess are considered rather easy these days. Captcha cracking, basic OCR, things I remember from my AI education at university. They are problems that are reasonably large and therefore don't really lend themselves to experimenting efficiently with different NN architectures.
As you probably know, Joel Grus came out with FizzBuzz in tensorflow. TLDR: learning from a binary representation of a number (ie. 12 bits encoding the number) into 4 bits (none_of_the_others, divisible by 3, divisible by 5, divisible by 15). For this toy problem, you can quickly compare different networks.
So I've been trying a simple feedforward network and wrote a program to compare various architectures. Things like a 2-hidden-layer feedforward network, then 3 layers, different activation functions, ... Most architectures, well, suck. They get somewhere near 50-60 success rate and remain there, independent of how much training you do.
A few perform really well. For instance, a sigmoid-activated double hidden layer with 23 neurons each works really well (89-90% correct after 2000 training epochs). Unfortunately anything close to it is rather disastrously bad. Take one neuron out of the second or first layer and it drops to 30% correct. Same for taking it out of the first layer ... Single hidden layer, 20 neurons tanh activated does pretty well as well. But most have a little over half this performance.
Now given that for real problems I can't realistically do these sorts of studies of different architectures, are there ways to get good architectures guaranteed to work ?
You might find the paper by Yoshua Bengio on Practical Recommendations for Gradient-Based Training of Deep Architectures helpful to learn more about hyperparameters and their settings.
If you're asking specifically for settings that have more guaranteed succes, I advise you to read on Batch Normalization. I find that it decreases the failure rate for bad picks of the learning rate and weight initialization.
Some people also discourage the use of non-linearities like sigmoid() and tanh() as they suffer from the vanishing gradient problem
Related
I know CNN has a lot of good features like weight sharing, save memory and feature extracting. However, this question makes me very confused. Is there any possible situation that fully connected network better than CNN? Why?
Thanks a lot guys!
Is there any possible situation that fully connected network better than CNN?
Well, I think we should first define what we mean by "better". Accuracy and precision are not the only things to consider: computational time, degrees of freedom and difficulty of the optimization should also be taken into account.
First, consider an input of size h*w*c. Feeding this input to a convolutional layer with F featuremaps and kernel size s will result in at about F*s*s*c learnable parameters (assuming there are no constraints on the ranks of the convolutions, otherwise we even have less parameters.). Feeding the same input into a fully connected layer with the same number of featuremaps will result in F*d_1*d_2*w*h*c, (where d_1,d_2 are the dimensions of each featuremap) which is clearly in the order of billions of learnable parameters given any input image with decent resolution.
While it can be tempting to think that we can get away with shallower networks (we already have lots of parameters, right?), fully connected layers are just linear layers after all, so we still need to insert many non-linearities in order for the network to gain reasonable representational power. So, this will mean that you will still need a deep network, however with so many parameters that it would be untractable. In addition, a larger network will have more degrees of freedom, and will therefore model much more than what we want: it will model noise unless we feed it some data or constrain it.
So yes, there might be a fully connected network that in theory could give us better performance, but we don't know how to train it yet. Finally, and this is purely based on intuition and therefore might be wrong, but it seems unlikely to me that such a fully connected network would converge to a dense solution. Since many convolutional networks achieve very high levels of accuracy (99% and up) on many tasks, I think that the optimal solution the fully connected network would converge to would be close to the convolutional network. So, we don't really need to train the fully connected one, but just a subset of its architecture.
I have a use case where I have around 100 images each of 10000 unique items. I have 10 items with me which are all from the 10000 set and I know which 10 items too but only at the time of testing on live data. I have to now match the 10 items with their names. What would be an efficient way to recognise these items? I have full control of training environment background and the testing environment background. If I make one model of all 10000 items, will it scale? Or should I make 10000 different models and run the 10 items on the 10 models I have pretrained.
Your question is regarding something called "one-vs-all classification" you can do a google search for that, the first hit is a video lecture by Andrew Ng that's almost certainly worth watching.
The question has been long studied and in a plethora of contexts. The answer to your question does very much depend on what model you use. But I'll assume that, if you're doing image classification, you are using convolutional neural networks, because, after all, they're state of the art for most such image classification tasks.
In the context of convolutional networks, there is something called "Multi task learning" that you should read up on. Boiled down to a single sentence, the concept is that the more you ask the network to learn the better it is at the individual tasks. So, in this case, you're almost certain to perform better training 1 model on 10,000 classes than 10,000 classes each performing a one-vs-all classification scheme.
Take for example the 1,000 class Imagenet dataset, and CIFAR-10's 10 class dataset. It has been demonstrated in numerous papers that first training against Imagenet's 1,000 class dataset, and then simply replacing the last layer with a 10 class output and re-training on CIFAR-10's dataset will produce a better result than just training on CIFAR-10's dataset alone. There are admittedly multiple reasons for this result, Imagenet is a larger dataset. But the richness of class labels, multi-task learning, in the Imagenet dataset is certainly among the reasons for this result.
So that was a long winded way of saying, use one model with 10,000 classes.
An aside:
If you want to get really, really interesting, and jump into the realm of research level thinking, you might consider a 1-hot vector of 10,000 classes rather sparse and start thinking about whether you could reduce the dimensionality of your output layer using an embedding. An embedding would be a dense vector, let's say size 100 as a good starting point. Now class labels turn into clusters of points in your 100 dimensional space. I bet your network will perform even better under these conditions.
If this little aside didn't make sense, it's completely safe to ignore it, your 10,000 class output is fine. But if it did peek your interest look up information on Word2Vec, and read this really nice post on how face recognition is achieved using embeddings: https://medium.com/#ageitgey/machine-learning-is-fun-part-4-modern-face-recognition-with-deep-learning-c3cffc121d78. You might also consider using an Auto Encoder to generate an embedding for the images (though I favor triplet embeddings as typically used in face recognition myself).
I have faced some problem when I needed to solve Regression Task and use as minimum instances as possible. When I tried to use Xgboost I had to feed 4 instances to get the reasonable result. But Multilayer Perceptron tuned to overcoming Regression problems has to take 20 instances, tried to change amount of neurons&layers but the answer is still 20 .Is it possible to do something to make Neural Network solve Resgression tasks with from 2 to 4 instances? if yes - explain please what should I do to succeed in it? Maybe there is some correlation between how much instances are needed to train and get reasonable results from Perceptron and how features are valuable inside dataset?
Thanks in advance for any help
With small numbers of samples, there are likely better methods to apply, Xgaboost definitely comes to mind as a method that does quite well at avoiding overfitting.
Neural networks tend to work well with larger numbers of samples. They often over fit to small datasets and underperform other algorithms.
There is, however, an active area of research in semi-supervised techniques using neural networks with large datasets of unlabeled data and small datasets of labeled samples.
Here's a paper to start you down that path, search on 'semi supervised learning'.
http://vdel.me.cmu.edu/publications/2011cgev/paper.pdf
Another area of interest to reduce overfitting in smaller datasets is in multi-task learning.
http://ruder.io/multi-task/
Multi task learning requires the network to achieve multiple target goals for a given input. Adding more requirements tends to reduce the space of solutions that the network can converge on and often achieves better results because of it. To say that differently: when multiple objectives are defined, the parameters necessary to do well at one task are often beneficial for the other task and vice versa.
Lastly, another area of open research is GANs and how they might be used in semi-supervised learning. No papers pop to the forefront of my mind on the subject just now, so I'll leave this mention as a footnote.
What are the potential reasons for a NN to output different values for the same input? Especially when there isn't any random or stochastic processes?
This is a very broad and general question, might be even too broad to even be on here, but there are several things you should know about neural networks:
They are NOT methods for finding one prefect optimal solution. A neural network usually learn examples that it is given and "figures out" a way to predict results reasonably well. Reasonable is relative, and for some models may mean 50% success and for others anything short of 99.9% will be considered failure.
They're outcome is very dependent on the data that was trained on. The order of data matters, and it's usually a good idea to shuffle data during training, but that can lead to wildly different results. Also, the quality of data matters - if the training data is very different in nature to the test data for example.
The best analogy of neural networks in computing is of course - the brain. Even with the same information and same basic underlying biology, we could all evolve different opinions on matters based on endless other variables. Same thing with computer learning to some extent.
Some types of neural networks use dropout layers, that are specifically designed to shut off random parts of the network during training. This should not affect the final prediction process, because for predictions that layer is usually set to allow all the parts of the network to operate, but if you are inputting data and telling the model it is "training" instead of asking it to predict, the results may vary significantly.
The sum of all this is just to say: The training of neural networks should be expected to yield different results from similar starting conditions, and so must be tested multiple times for every condition to determine what parts of it are inevitable and what parts are not.
It might be due to shuffling of data , If you want to use the same vector you should turn the shuffle argument off.
You should try disabling dropout. Dropout randomly sets the outputs of certain neurons to 0. This will mean that your output will be different each time.
This is probably a newbie question but I'm trying to get my head around how training on small batches works.
Scenario -
For the mnist classification problem, let's say that we have a model with appropriate hyerparameters that allow training on 0-9 digits. If we feed it with a small batches of uniform distribution of inputs (that have more or less same numbers of all digits in each batch), it'll learn to classify as expected.
Now, imagine that instead of a uniform distribution, we trained the model on images containing only 1s so that the weights are adjusted until it works perfectly for 1s. And then we start training on images that contain only 2s. Note that only the inputs have changed, the model and everything else has stayed the same.
Question -
What does the training exclusively on 2s after the model was already trained exclusively on 1s do? Will it keep adjusting the weights till it has forgotten (so to say) all about 1s and is now classifying on 2s? Or will it still adjust the weights in a way that it remembers both 1s and 2s?
In other words, must each batch contain a uniform distribution of different classifications? Does retraining a trained model in Tensorflow overwrite previous trainings? If yes, if it is not possible to create small (< 256) batches that are sufficiently uniform, does it make sense to train on very large (>= 500-2000) batch sizes?
That is a good question without a clear answer. In general, the order and selection of training samples has a large impact on the performance of the trained net, in particular in respect to the generalization properties it shows.
The impact is so strong, actually, that selecting specific examples, and ordering them in a particular way to maximize performance of the net even constitutes a genuine research area called `curriculum learning'. See this research paper.
So back to your specific question: You should try different possibilities and evaluate each of them (which might actually be an interesting learning exercise anyways). I would expect uniformly distributed samples to generalize well over different categories; samples drawn from the original distribution to achieve the highest overall score (since, if you have 90% samples from one category A, getting 70% over all categories will perform worse than having 99% from category A and 0% everywhere else, in terms of total accuracy); other sample selection mechanisms will show different behavior.
An interesting reading about such questions is Bengio's 2012 paper Practical Recommendations for Gradient-Based Training of Deep
Architectures
There is a section about online learning where the distribution of training data is unknown. I quote from the original paper
It
means that online learners, when given a stream of
non-repetitive training data, really optimize (maybe
not in the optimal way, i.e., using a first-order gradient
technique) what we really care about: generalization
error.
The best practice though to figure out how your dataset behaves under different testing scenarios would be to try them both and get experimental results of how the distribution of the training data affects your generalization error.