User Kari

14 votes

Accepted

Q-Learning: Target Network vs Double DQN

answered May 29, 2018 at 22:00

9 votes

Why do we need 2 matrices for word2vec or GloVe

answered Mar 16, 2018 at 14:55

8 votes

Accepted

How does Implicit Quantile-Regression Network (IQN) differ from QR-DQN?

answered Nov 10, 2018 at 15:15

6 votes

Accepted

Q learning - how to use experience replay, when playing against other agent?

answered Feb 3, 2018 at 2:54

4 votes

Why is my test data accuracy higher than my training data?

answered May 23, 2019 at 16:40

3 votes

Accepted

Nesterov Accelerated gradient is in correct order?

answered Jan 8, 2018 at 1:02

3 votes

How backpropagation through gradient descent represents the error after each forward pass

answered Dec 10, 2017 at 4:14

2 votes

Accepted

My LSTM can't reduce error down to zero, when overfitting

answered Dec 12, 2017 at 2:12

2 votes

Accepted

Synthetic Gradients - what's the practical benefit?

answered May 31, 2018 at 1:01

2 votes

How to determine if problem is solvable by neural network?

answered Mar 21, 2018 at 16:31

2 votes

Accepted

Neural Turing Machine - how the shifting works?

answered Mar 22, 2018 at 13:18

2 votes

Accepted

How much neural network theory required to design one?

answered Jun 16, 2018 at 20:03

2 votes

Accepted

RNNs for time series prediction - what configurations would make sense

answered Jun 28, 2018 at 19:02

2 votes

Hindsight Experience Replay, how to define a partially-known End-Goal

answered Jul 2, 2018 at 22:54

1 vote

LSTM training/prediction with no starting sequence

answered Aug 19, 2018 at 22:34

1 vote

Variable time units LSTM

answered Oct 30, 2018 at 21:57

1 vote

More layers in NN give worse result

answered Jun 6, 2018 at 23:24

1 vote

Accepted

Prioritized Experience Replay - why to approximate the Density Function?

answered Jun 10, 2018 at 1:41

1 vote

Accepted

LSTM: Taking previous output values as feature

answered Dec 27, 2017 at 23:24

1 vote

Accepted

use hidden layer of word2vec instead of 'one-hot', to reduce number of weights in other nets?

answered Mar 9, 2018 at 3:37

1 vote

Synthetic Gradients good number of Layers & neurons

answered Dec 14, 2018 at 12:04

1 vote

Accepted

recurrent weights on LSTM Data-gate? (the one with tanh)

answered Nov 10, 2018 at 22:32

1 vote

Accepted

Dueling Network gradient with respect to Advantage stream

answered Jun 18, 2019 at 21:29

1 vote

Neural Network backprop formula - Matrix dimensions won't match?

answered Jun 29, 2019 at 10:19

1 vote

Deep Q-Learning for physical quantity: q-values distribution not as expected

answered Jul 29, 2019 at 23:39

1 vote

Getting Validation Accuracy of 99% with MNIST with less than 10000 parameters CNN

answered Sep 15, 2019 at 12:22

0 votes

Accepted

Gradient Checking: MeanSquareError. Why huge epsilon improves discrepancy?

answered Apr 25, 2020 at 20:59

0 votes

Standarizing the Label to postive intergers in Machine Learning Classification task, why is it recommended?

answered Feb 6, 2022 at 15:44

0 votes

Muti-hot encoding vs Label-Encoding

answered Nov 15, 2018 at 23:15

0 votes

Accepted

Gradient Checking LSTM - how to get change in Cost across timesteps?

answered Nov 22, 2018 at 15:38

Stack Exchange Network

33 Answers

Q-Learning: Target Network vs Double DQN

Why do we need 2 matrices for word2vec or GloVe

How does Implicit Quantile-Regression Network (IQN) differ from QR-DQN?

Q learning - how to use experience replay, when playing against other agent?

Why is my test data accuracy higher than my training data?

Nesterov Accelerated gradient is in correct order?

How backpropagation through gradient descent represents the error after each forward pass

My LSTM can't reduce error down to zero, when overfitting

Synthetic Gradients - what's the practical benefit?

How to determine if problem is solvable by neural network?

Neural Turing Machine - how the shifting works?

How much neural network theory required to design one?

RNNs for time series prediction - what configurations would make sense

Hindsight Experience Replay, how to define a partially-known End-Goal

LSTM training/prediction with no starting sequence

Variable time units LSTM

More layers in NN give worse result

Prioritized Experience Replay - why to approximate the Density Function?

LSTM: Taking previous output values as feature

use hidden layer of word2vec instead of 'one-hot', to reduce number of weights in other nets?

Synthetic Gradients good number of Layers & neurons

recurrent weights on LSTM Data-gate? (the one with tanh)

Dueling Network gradient with respect to Advantage stream

Neural Network backprop formula - Matrix dimensions won't match?

Deep Q-Learning for physical quantity: q-values distribution not as expected

Getting Validation Accuracy of 99% with MNIST with less than 10000 parameters CNN

Gradient Checking: MeanSquareError. Why huge epsilon improves discrepancy?

Standarizing the Label to postive intergers in Machine Learning Classification task, why is it recommended?

Muti-hot encoding vs Label-Encoding

Gradient Checking LSTM - how to get change in Cost across timesteps?