Is there a way to use RNN (in tensorflow) to do something like a batch Kalman with the weight dynamics specified in the loss?

Or would you simply do this as a time series of models. Basically I think you can think of time series of weights as the hidden states and the dynamics driving the weight time series as the RNN weights. I’m not sure if the data gradients are avoiding look ahead in this context though. I…

Performance of the skip-gramm(word2vec) model for the words similarity using brown dataset(NLTK)

I want to create similarity matrix based on the brown dataset from the NLTK library. The problem is that loss tf.reduce_mean(tf.nn.sampled_softmax_loss(weights = softmax_weight, biases = softmax_bias, inputs = embed, labels = y, num_sampled = num_sampled, num_classes = num_words)) decreases from 4.2 to 2.0 and then it starts to go up and down. The question is:…

Why does randomizing samples of reinforcement learning model with a non-linear function approximator reduce variance?

I have read the DQN thesis. While reading the DQN paper, I found that randomly selecting and learning samples reduced divergence in RL using a non-linier function approximator. If so, why is the learning of RL using a non-linier function approximator divergent when the input data are strongly correlated?

feasible to use GAN for high quality image synthesis other than human face?

The famous nvidia paper can generate hyperrealistic human face. But in the very same paper https://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf images of other categories are rather disappointing and there hasn’t seemed to be any improvements since then. I wonder why it is the case? is it because they didn’t have enough training data for other categories? Or is it…