Large word embedding matrix update in tensorflow - machine-learning

Recently I am using a CNN to do text classification which is described in http://www.wildml.com/2015/12/implementing-a-cnn-for-text-classification-in-tensorflow/.
My dataset is quite large and the vocabulary size is more than 1M words. My training algorithm becomes much slower when the vocabulary size gets bigger. There is a warning message saying that "....tensorflow/python/ops/gradients.py:87: UserWarning: Converting sparse IndexedSlices to a dense Tensor with 145017088 elements. This may consume a large amount of memory."
I am thinking this may be caused by a dense gradient update on the embedding matrix. Does anyone have any ideas on that? There is a post Update only part of the word embedding matrix in Tensorflow discussing similar issue. The top answer states that tensorflow only updates specific rows. Why did it change in my case?
Thanks!

Related

Is there a fundamental limit on how accurate location information is encoded in CNNs

Each layer in a CNN reduces the size of the input via convolution and max-pooling operations. Convolution is translation equivariant, but max-pooling is translation invariant. Correct me if this is wrong : each time max-pooling applied, the precise location of a feature is reduced. So the feature maps of the final conv layer in a very deep CNN will have a large receptive field (w.r.t the original image), but the location of this feature (in the original image) is not discernible from looking at this feature map alone.
If this is true, how can the accuracy of bounding boxes when we do localisation be so good with a deep CNN? I understand how classification works, but making accurate bounding box predictions is confusing me.
Perhaps a toy example will clarify my confusion;
Say we have a dataset of images with dimension 256x256x1, and we want to predict whether a cat is present, and if so, where it is, so our target is something like [sigmoid_cat_present, cat_location].
Our vanilla CNN (let's assume something like VGG) will take in the image and transform it to something like 16x16x256 in the last convolutional layer. Each pixel in this final 16x16 feature map can be influenced by a much larger region in the original image. So if we determine a cat is present, how can the [cat_location] be refined to value more granular than this effective receptive field?
To add to your question - how about pixel perfect accuracy of segmentation boundary !!
Your intuition regarding down-sampling via max-pooling is correct. Normal CNNs have that limit. However, there have been some improvements recently to overcome it.
The breakthrough to this problem came in 2015-6 in the form of U-net and atrous/dilated convolution introduced in DeepLab.
Dilated convolutions or atrous convolutions, previously described for wavelet analysis without signal decimation, expands window size without increasing the number of weights by inserting zero-values into convolution kernels. Dilated convolutions have been shown to decrease blurring in semantic segmentation maps, and are purported to work at least in part by extracting long range information without the need for pooling.
Using U-Net architectures is another method that seeks to retain high spatial frequency information by directly adding skip connections between early and late layers. In other words, up-sampling followed by down-sampling.
In TensorFlow, atrous convolutions are implemented with function:
tf.nn.atrous_conv2d
There are many more methods and this is an ongoing research area.

Tighter clustering on a larger dataset?

not sure if im asking this question in the correct place but anyway...
I have been using sci-kit learn package to perform dimensionality reduction on 2 different datasets, one is a large dataset with several thousand genes and the other is a smaller dataset with only the most relevant 50 genes as identified as professionals in the area my data is centered around.
Basically with nearly all dimensionality reduction techniques I am seeing that the smaller dataset gives a higher accuracy (as would be expected).
However whenever I perform LDA(Linear Discriminatory Analysis) I am seeing the opposite effect where the larger dataset has a higher accuracy (tighter clusters, higher purity and higher accuracy using KNN).
Do you guys have advice on ways I could check that I am not making an error or ways to investigate why this is happening?
Thanks :)

Reducing pixels in large data set (sklearn)

Im currently working on a classification project but I'm in doubt about how I should start off.
Goal
Accurately classifying pictures of size 80*80 (so 6400 pixels) in the correct class (binary).
Setting
5260 training samples, 600 test samples
Question
As there are more pixels than samples, it seems logic to me to 'drop' most of the pixels and only look at the important ones before I even start working out a classification method (like SVM, KNN etc.).
Say the training data consists of X_train (predictors) and Y_train (outcomes). So far, I've tried looking at the SelectKBest() method from sklearn for feature extraction. But what would be the best way to use this method and to know how many k's I've actually got to select?
It could also be the case that I'm completely on the wrong track here, so correct me if I'm wrong or suggest an other approach to this if possible.
You are suggesting to reduce the dimension of your feature space. That is a method of regularization to reduce overfitting. You haven't mentioned overfitting is an issue so I would test that first. Here are some things I would try:
Use transfer learning. Take a pretrained network for image recognition tasks and fine tune it to your dataset. Search for transfer learning and you'll find many resources.
Train a convolutional neural network on your dataset. CNNs are the go-to method for machine learning on images. Check for overfitting.
If you want to reduce the dimensionality of your dataset, resize the image. Going from 80x80 => 40x40 will reduce the number of pixels by 4x, assuming your task doesn't depend on fine details of the image you should maintain classification performance.
There are other things you may want to consider but I would need to know more about your problem and its requirements.

CNN Regression on Grid - Limitation of Convolutional Neural Networks?

I'm working on a (high energy physics related) problem using CNNs.
For understanding the problem, let's consider these examples here.
The left-hand side is the input to the CNN, the right-hand side the desired output. So the network is supposed to cluster the input. The actual algorithm behind this clustering (i.e. how we got the desired output for training) is really complex and we want the CNN to learn this.
I've tried different CNN architectures, for example one similar to the U-net architecture (https://arxiv.org/abs/1505.04597) but also various concatenations of convolutional layers, etc.
The outputs are always really similar (for all architectures).
Here you can see some CNN predictions.
In principle the network is performing quite well, but as you can see, in most cases the CNN output consists of several filled pixels that are directly next to each other, which will never (!) happen in the true cases.
I've been using mean squared error as the loss function in all of the networks.
Do you have any suggestions how one could avoid this problem and improve the networks performance?
Or is this a general limitation to CNNs and in practice it is not possible to solve such a problem using CNNs?
Thank you very much!
My suggestion would be to split up the work. First use a U-Shaped NN to find the activations in a binary segmentation task (like in your paper) and then regress on the found activations to find their final values. In my experience this works way better than doing regression on large images, because the MSE will result in blurry outputs, as you have observed.
The CNN does not know that you wanted a sharp result. As mentioned by #Thomas, MSE tends to give you blurry result as it is the nature of that loss function. Giving a blurry result does not introduce large loss in MSE.
An easy modification would be to use L1 Loss (absolute difference instead of squared error). It has a constant gradient unlike MSE whose gradient decreases with error.
If you really wanted a sharp result, it would be easier to add a manual step -- non maximum suppression (NMS). In practice, a 3x3 box-max filter might do.

Possible/maybe category in deep learning

I'm interested in taking advantage of some partially labeled data that I have in a deep learning task. I'm using a fully convolutional approach, not sampling patches from the labeled regions.
I have masks that outline regions of definite positive examples in an image, but the unmasked regions in the images are not necessarily negative - they may be positive. Does anyone know of a way to incorporate this type of class in a deep learning setting?
Triplet/contrastive loss seems like it may be the way to go, but I'm not sure how to accommodate the "fuzzy" or ambiguous negative/positive space.
Try label smoothing as described in section 7.5.1 of Deep Learning book:
We can assume that for some small constant eps, the training set label y is correct with probability 1 - eps, and otherwise any of the other possible labels might be correct.
Label smoothing regularizes a model based on a softmax with k output values by replacing the hard 0 and 1 classification targets with targets of eps / k and 1 - (k - 1) / k * eps, respectively.
See my question about implementing label smoothing in Pandas.
Otherwise if you know for sure, that some areas are negative, other are positive while some are uncertain, then you can introduce a third uncertain class. I have worked with data sets that contained uncertain class, which corresponded to samples that could belong to any of the available classes.
I'm assuming that you are struggling with a data segmantation task with a problem of a ill-definied background (e.g. you are not sure if all examples are correctly labeled). Recently I came across the similiar problem and this is what I came across during my research:
In old days before deep learning and at the begining of deep learning era - the common way to deal with that is to smooth your output with some kind of a probability model which would take into account the possibility of a noisy labels (you could read about this in a Learning to Label from Noisy Data chapter from this book. It's important to discriminate this probabilistic models from models used to smooth your labels w.r.t. to image or label structure like classical CRFs for bilateral smoothing.
What we finally used (and worked really well) is the Channel Inhibited Softmax idea from this paper. In terms of a mathematical properties - it makes your network much more robust to some objects not labeled - because it makes your network to output much higher positive valued logits at correctly labeled objects.
You could treat this as a semi-supervised problem. Use the full dataset without labels to train a bottleneck autoencoder structure (or a GAN approach). This pretrained model can then be adjusted (e.g. removing the last layers, adding a better layer structure at the end on top of the bottleneck features) and finetuned on the labeled data.

Resources