How to implement skip connections from MSG-GAN paper - tensorflow

I am trying to implement the technique described in the MSG-GAN paper:
https://arxiv.org/pdf/1903.06048.pdf
But I am having difficulty understanding some things, for example, how are the connections made from the generator to the discriminator? These are Conv2D connections literally? (in that case, how would I insert the real images to train the discriminator?) Or does the discriminator have multiple outputs (one prediction for each resolution and the generator has to optimize the average loss of the resolutions)?

How are the connections made from the generator to the discriminator?
Generator output them and discriminator concatenate them with feature maps from last layer at corresponding input layer.
These are Conv2D connections literally?
Those just input tensor with shape like(batch size, W, H, 3), same as ordinary image input.
Does the discriminator have multiple outputs?
No, this is end to end training, training with all resolution outputs at same time, otherwise it will just like the Progressive Growing GAN and no reason for concatenate operation at each input layer(beginning of each block) of discriminator.

this is only a partial answer.
I would say that if you had to implement this in keras, and you don't want each model (G and D) to be in one piece, it's actually easier to have them separated and then use tf.GradientTape() to train
Does the discriminator have multiple outputs?
yes, if implementing them as separate models, yes there will be multiple inputs and multiple outputs of multiple resolutions, only one of those is the final output.

Related

Multiple BERT binary classifications on a single graph to save on inference time

I have five classes and I want to compare four of them against one and the same class. This isn't a One vs Rest classifier, as for each output I want to score them against one base class.
The four outputs should be: base class vs classA, base class vs classB, etc.
I could do this by having multiple binary classification tasks, but that's wasting computation time if the first layers are BERT preprocessing + pretrained BERT layers, and the only differences between the four classifiers are the last few layers of BERT (finetuned ones) and the Dense layer.
So why not merge the graphs for more performance?
My inputs are four different datasets, each annotated with true/false for each class.
As I understand it, I can re-use most of the pipeline (BERT preprocessing and the first layers of BERT), as those have shared weights. I should then be able to train the last few layers of BERT and the Dense layer on top differently depending on the branch of the classifier (maybe using something like keras.switch?).
I have tried many alternative options including multi-class and multi-label classifiers, with actual and generated (eg, machine-annotated) labels in the case of multiple input labels, different activation and loss functions, but none of the results were acceptable to me (none were as good as the four separate models).
Is there a solution for merging the four different models for more performance, or am I stuck with using 4x binary classifiers?
When you train DNN for specific task it will be (in vast majority of cases) be better than the more general model that can handle several task simultaneously. Saying that, based on my experience the properly trained general model produces very similar results to the original binary ones. Anyways, here couple of suggestions for training strategies (assuming your training datasets for each task are completely different):
Weak supervision approach
Train your binary classifiers, and label your datasets using them (i.e. label with binary classifier trained on dataset 2 datasets [1,3,4]). Then train your joint model as multilabel task using all the newly labeled datasets (don't forget to randomize samples before feeding them to trainer ;) ). Here you will need to experiment if you will use threshold and set a label to 0/1 or use the scores of the binary classifiers.
Create custom loss function that will not penalize if no information provided for certain class. So when your will introduce sample from (say) dataset 2, your loss will be calculated only for the 2nd class.
Of course you can apply both simultaneously. For example, if you know that binary classifier produces scores that are polarized (most results are near 0 or 1), you can use weak labels, and automatically label your data with scores. Now during the second stage penalize loss such that for score x' = 4(x-0.5)^2 (note that you get logits from the model, so you will need to apply sigmoid function). This way you will increase contribution of the samples binary classifier is confident about, and reduce that of less certain ones.
As for releasing last layers of BERT, usually unfreezing upper 3-6 layers is enough. Releasing more layers improves results very little and increases time and memory requirements.

Limiting probability percentage of irrelevant image in CNN

I am training a cnn model with five classes using keras library. Using model.predict function i get prediction percentage of the classes. My problem is for a image which doesn't belong to these classes and completely irrelevant, the predict class still predicts the percentages according to the classes.
How do I prevent it? How do I identify it as irrelevant?
I assume you are using a softmax activation on your last layer to generate the probabilities for each class. By definition, the sum of the outputs from the softmax activation must add up to 1. Therefore, it is impossible for the neural net to say that the image does not belong to any of your classes, with your current setup.
There are two potential ways you could address this:
Add another class that represents "other" or "unknown" objects (so you have 6 classes).
Add another output to your neural net (or train a completely independent neural net) that does binary classification on whether or not the image is in one of the 5 classes. That way, if your secondary output says that the image is not in the 5 classes, you can ignore the softmax output.
In both cases, you will need to augment your dataset with images that do not fall in your 5 classes.

Convolutional Neural Network Training

I have a question regarding convolutional neural network (CNN) training.
I have managed to train a network using tensorflow that takes an input image (1600 pixels) and output one of three classes that matches it.
Testing the network with variations of the trained classes is giving good results. However; when I give it a different -fourth- image (does not contain any of the trained 3 image), it always returns a random match to one of the classes.
My question is, how can I train a network to classify that the image does not belong to either of the three trained images? A similar example, if i trained a network against the mnist database and then a gave it the character "A" or "B". Is there a way to discriminate that the input does not belong to either of the classes?
Thank you
Your model will always make predictions like your labels, so for example if you train your model with MNIST data, when you will make predictions, prediction will always be 0-9 just like MNIST labels.
What you can do is train a different model first with 2 classes in which you will predict if an image belongs to data set A or B. E.x. for MNIST data you label all data as 1 and add data from other sources that are different (not 0-9) and label them as 0. Then train a model to find if image belongs to MNIST or not.
Convolutional Neural Network (CNN) predicts the result from the defined classes after training. CNN always return from one of the classes regardless of accuracy. I have faced similar problem, what you can do is to check for accuracy value. If the accuracy is below some threshold value then it's belong to none category. Hope this helps.
You probably have three output nodes, and choose the maximum value (one-hot encoding). That's a bit unfortunate as it's a low number of outputs. Non-recognized inputs tend to cause pretty random outputs.
Now, with 3 outputs, roughly speaking you can get 7 outcomes. You might get a single high value (3 possibilities) but non-recognized input can also cause 2 high outputs (also 3 possibilities) or approximately equal output (also 3 possibilities). So there's a decent chance (~ 3/7) of random inputs producing a pattern on the output nodes which you'd only expect for a recognized input.
Now, if you had 15 classes and thus 15 output nodes, you'd be looking at roughly 32767 possible outcomes for unrecognized inputs, only 15 of which correspond to expected one-hot outcomes.
Underlying this is a lack of training data. If your training set has examples outside the 3 classes, you can just dump this in a 4th "other" category and train with that. This by itself isn't a reliable indication, as usually the theoretical "other" set is huge, but you now have 2 complementary ways of detecting other inputs: either by the "other" output node or by one of the 11 ambiguous outputs.
Another solution would be to check what outcome your CNN usually gives when given something else. I believe the last layer must be softmax and your CNN should return probabilities of the three given classes. If none of these probabilities is close to 1 this might be a sign that this is something else assuming your CNN is well trained (it must be fined for overconfidence when predicting wrong labels).

When predicting with an LSTM in Keras, is the hidden state still adjusted?

When I first train an LSTM in Keras on sequence data - my training data -
and then use model.predict() to make predictions with my test data as input, is the hidden state of the LSTM still being adjusted?
Basic operation of a neural network is to take an input (vector) which is connected to the output with connections and, sometimes, other layers such as context layers. These connections are modelled as matrices and vary in strength, we call these weight matrices.
This means that the only thing we do when we are feeding data into the network is to put a vector into the network, multiply the values with the weight matrix and call that the output. In special cases, like recurrent networks, we even keep some values stored in other vectors and combine this stored value with the current input.
During training we not only feed data into the network, we also compute an error value that we evaluate in a clever way so that it tells us how we should change the weight matrices we multiply our inputs (and possibly past inputs for recurrent layers) with.
Therefore: yes, of course the basic execution behavior does not change for recurrent layers. We are just not updating weights anymore.
There are layers that do behave differently during execution time because they are treated as regularisers, i.e. methods that make training the network more efficient, which are deemed as unnecessary during execution. Examples for these layers are Noise and BatchNormalization. Almost all neural network layers (including recurrent ones) include drop-out which is another form of regularisation which disables a random percentage of connections in the layer. This is also only done during training.

GAN with not a random input

I'm very interested in GAN those times.
I coded one for MNIST with the following structure :
Generator model
Discriminator model
Gen + Dis model
Generator model generate batches of image from random distribution.
Discrimator is trained over it and real images.
Then Discriminator is freeze in Gen+Dis model and Generator trained. (With the frozen Discriminator who says if the generator is good or not)
Now, imagine I don't want to feed my generator with a random distribution but with images. (For upscaling for example, or generate an real image from a draw)
Do I need to change something in it ?
(Except the conv model who will be more complex)
Should I continue to use the binary_crossentropy as loss function ?
Thanks you very much!
You can indeed put a variational autoencoder (VAE) in front in order to generate the initial distribution z (see paper).
If you are interested in the topic I can recommend the this course at Kadenze.