User Hugh Perkins - Cross Validated

Hugh Perkins

Member for 10 years, 8 months
Last seen more than a week ago

GitHub
linkedin.com/in/hughperkins
New York, NY, United States

106 Answers

Score Activity Newest

30 votes

What are "residual connections" in RNNs?

answered Jan 1, 2018 at 21:01

23 votes

Accepted

Gradient backpropagation through ResNet skip connections

answered Mar 21, 2017 at 8:42

18 votes

Accepted

How to train LSTM model on multiple time series data?

answered Jul 25, 2018 at 1:56

15 votes

Accepted

Data augmentation on training set only?

answered Dec 31, 2017 at 3:00

13 votes

Accepted

Can a standard deviation of raw scores be reported as a standard deviation of percentages?

answered Oct 4, 2017 at 10:49

11 votes

What is the reason that the Adam Optimizer is considered robust to the value of its hyper parameters?

answered Oct 9, 2017 at 20:35

10 votes

Modern Use Cases of Restricted Boltzmann Machines (RBM's)?

answered Oct 9, 2017 at 20:46

10 votes

What is the difference between regular PCA and probabilistic PCA?

answered Oct 9, 2017 at 20:56

9 votes

What is the root cause of the class imbalance problem?

answered Mar 18, 2017 at 23:19

8 votes

need intuition: why is there no winning strategy to this game?

answered Mar 26, 2017 at 21:01

7 votes

How to interpret the results when both ridge and lasso separately perform well but produce different coefficients

answered Mar 14, 2017 at 12:44

6 votes

Accepted

How do we incorporate new information into a Dirichlet prior distribution?

answered Oct 9, 2017 at 14:17

6 votes

Accepted

How to distinguish episodic task and continuous tasks?

answered Apr 2, 2017 at 8:50

5 votes

Accepted

Interpret predictions of black box models

answered Mar 27, 2017 at 12:36

5 votes

Accepted

LSTM clarification on output

answered Mar 26, 2017 at 13:24

5 votes

What are the benefits of using ReLU over softplus as activation functions?

answered Mar 16, 2017 at 10:03

5 votes

deep learning - word embedding with parts of speech

answered Feb 12, 2018 at 0:08

4 votes

How to find derivative of softmax function for the purpose of gradient descent?

answered Feb 12, 2018 at 0:28

4 votes

Accepted

In the Metropolis algorithm, if the ratio of probabilities $r$ is less than $1$, why not directly reject instead of accepting with probability $r$?

answered Mar 23, 2017 at 23:51

4 votes

Accepted

How to train a CNN with multiple heads?

answered Mar 24, 2017 at 12:10

4 votes

Accepted

What does alignment between input and output mean for recurrent neural network

answered Apr 5, 2017 at 12:32

4 votes

When is $\mathbb{E}\left[\frac{1}{\sum X_{i}}\right] = \frac{1}{\mathbb{E}\left[\sum X_{i}\right]}$?

answered Oct 16, 2017 at 8:01

3 votes

Accepted

If z is uniform normal is $f(z;\theta)$ a normal distribution

answered Nov 25, 2017 at 15:45

3 votes

Accepted

What are nodes in RNN/LSTM?

answered Mar 27, 2017 at 8:52

3 votes

Accepted

Assumptions on class priors in expectation-maximization?

answered Mar 27, 2017 at 8:55

3 votes

MCMC: invalid covariance matrix due to numerical error

answered Apr 2, 2017 at 7:44

3 votes

Restricted Boltzmann Machine : how is it used in machine learning?

answered Mar 18, 2017 at 22:13

3 votes

For symmetric matrices, is the Cholesky decomposition better than the SVD?

answered Feb 11, 2018 at 16:45

3 votes

Accepted

Sequence to Sequence models: Why is the start sequence (e.g. "</s>") required?

answered Jan 2, 2018 at 2:34

3 votes

What problem does Residual Nets solve that batch normalization does not solve?

answered Dec 28, 2017 at 0:52

1