Community Profile

photo

Greg Heath


3 090 total contributions since 2011

Backgound in Electromagnetic Theory, Plasma Physics and Radar Target Identification using Neural Networks.
PhD Student, Research Assistant and Lecturer at Stanford;
AB, ScB, ScM Student; Research Assistant, Fellow and Professor at Brown;
27 yrs researching Ballistic and Theatre Missile Defense using Neural Networks at MIT Lincoln Laboratory. Retired 2003.

PLEASE DO NOT SEND QUESTIONS AND DATA TO MY EMAIL. HOWEVER, CAN SEND LINKS TO POSTS.
Professional Interests: Neural Netwoks, Spectral Analysis

Contact

Greg Heath's Badges

  • 36 Month Streak
  • Thankful Level 4
  • Ace
  • Revival Level 4
  • Knowledgeable Level 4
  • First Answer

View details...

Contributions in
View by

Answered
Is the "patternnet" a fully connected neural network
Yes. The only difference between my classifiers and regressors is the sigmoid output layer instead of linear. Hope this helps....

ungefär 3 timmar ago | 0

Answered
How to train and test time series data in Neural Network Toolbox
I order to test a net you have to compare the actual output with the desired output. Hope this helps. THANK YOU FOR FORMALLY...

3 dagar ago | 0

Answered
Artificial Neural Network implementation and to know the importance of each of the input on output(Response) - wanted help
The way I determine the importance of a single input is 1. Calculate the error using all inputs 2. Loop over inputs ...

3 dagar ago | 0

| accepted

Answered
transfer function purelin equation based on neural network toolbox
Read the documentation help purelin doc purelin Greg

12 dagar ago | 0

Answered
How to use a sequenceInputLayer with a regressionLayer (neural networks) ?
The answer is obvious: help regressionlayer doc regressionlayer *Thank you for formally accep...

19 dagar ago | 0

Answered
Time Domain Signal for neural network
If you have rpm I don't see why time is important. However, there is no reason you cannot do both and see what difference it ma...

19 dagar ago | 0

| accepted

Answered
Adding hidden layers to a patternnet hurts accuracy?
The global minimum is achievable with a single hidden layer. With more hidden layers you add more local minima; most of which ...

20 dagar ago | 0

| accepted

Answered
Training neural network - Forward Pass
You are very mixed up. I suggest going to the library and finding a good elementary NN book. Obviously, the appropriate info is...

20 dagar ago | 0

Answered
Hello, I'm working with artificial neural network.
Three layers are sufficient: input/hidden/output The input layer is NOT a neuron layer. The number of input nodes is the dimens...

22 dagar ago | 0

| accepted

Answered
Why neural network gives negative output ?
How different is the new data (e.g., Mahalanobis distance)? If you know the true outputs, how do the error rates compare? If y...

22 dagar ago | 0

| accepted

Answered
Rsq from NMSE in NN
NMSE = mse(trnopdb-net(trnipdb))/MSE00 i.e., NO TRANSPOSES 2. Rsq = R^2 3. Yes. Use separate calculations fo...

22 dagar ago | 1

| accepted

Answered
HOW DO I SEARCH IN ANSWERS ???
Oh! It only opens up at the top of the page if the page is sufficiently wide. Since I often use large type and 2 pages per scre...

27 dagar ago | 0

| accepted

Answered
Interpolation using Neural Networks
Plot the data. Look at the plots Are there regions that are seasonal? Separate the data that is relevant to your problem Ca...

28 dagar ago | 0

| accepted

Answered
How to plot only few classes for confusion matrix?
Oviously, you have to subsample the original matrices. Greg

29 dagar ago | 0

Answered
Open loop Training performance and closed loop training performance are good but multi-step prediction is bad. Reason?
You are not considering information from the autocorrelation fuction. Hope this helps Greg

30 dagar ago | 0

Answered
Formula for two layer FFNN
y1 = b1 + IW1 * x y2 = b2 + LW2 * tanh( y1 ) y3 = b3 + LW3 * tanh( y2 ) = b3 + LW3 * tanh( b2 + LW2 * tanh( b1 + IW1 * x ...

ungefär en månad ago | 0

| accepted

Answered
Input and target have different number of sampel
Train and TTrain have to be transposed. Hope this helps. THANK YOU FOR FORMALLY ACCEPTING MY ANSWER Greg

ungefär en månad ago | 0

| accepted

Answered
How to calculate accuracy for neural network algorithms?
I normalize the mean-square-error MSE = mse(error) = mse(output-target) by the minimum MSE obtained when th...

ungefär en månad ago | 0

| accepted

Answered
Neural Network input and targets have different of samples
Transpose both matrices. Thank you for formally accepting my answer Greg

ungefär en månad ago | 0

| accepted

Answered
How these plots (Performance, Training state, Regression) shows the training performance? How to figure out the training rate from these plots?
Training data plots are useful. However, there is no indication of how good the net will perform on nontraining data (THE TRUE ...

ungefär en månad ago | 0

Answered
Validation check = 0 for traingdm
What you are worrying about is irrelevant. Your data is so good you don't even need a validation subset.The main purpose of a va...

ungefär en månad ago | 1

Answered
Expressing equation in terms of sin/cos
If you substitute your solutions into LHS and get the RHS, then it is possible. Greg

ungefär 2 månader ago | 0

Answered
How to decide window size for a moving average filter?
I'm very surprised that none of the previous responses mentioned 1. Determine characteristic self correlation lengths usi...

ungefär 2 månader ago | 0

Answered
finding optimal neural network architecture using genetic algorithms
0. The genetic approach is a waste of time. It takes too long. 1.Typically, a single hidden layer is sufficient. 2. Minimize t...

ungefär 2 månader ago | 0

Answered
Neuron Network input variables-Missing data
Sorry: You have to predict the missing data as best you can. Greg

ungefär 2 månader ago | 0

Answered
Why my network is not giving the desired output
Design(training+validation), test and new data should all have the same summary statistics BEFORE NORMALIZATION. This may requir...

2 månader ago | 0

| accepted

Answered
How to predict future responses y(t + 1) from the training of a narxnet network with past data of x (t) and y (t)? (NARXNET)
YOU DO NOT HAVE X and Y !!! YOU HAVE X and T where T = Ydesired Hope this helps Thank you for formally accepting m...

2 månader ago | 1

| accepted

Answered
Testing a Backpropagation Neural Network
You are probably OVERTRAINING AN OVERFIT NET OVERFITTING: Using more unknown hidden nodes than number ...

2 månader ago | 0

| accepted

Answered
The performance of hidden neurons
I think I misinterpreted the question. Now I think you mean when I increase the number of hidden nodes from 4 to 5 why do I star...

2 månader ago | 0

Answered
Crossentropy loss function - What is a good performance goal?
These equations are not necessarily precise. For example: data = design + test design = training + validation In partic...

2 månader ago | 0

Load more