3,095 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

Answered

I get a "Performance function replaced with squared error performance" warning when trying to set 'crossentropy' as the performance function.

If you insist on using CROSSENTROPY, try PATTERNNET. Hope this helps. Thank you for formally accepting my answer Greg

I get a "Performance function replaced with squared error performance" warning when trying to set 'crossentropy' as the performance function.

If you insist on using CROSSENTROPY, try PATTERNNET. Hope this helps. Thank you for formally accepting my answer Greg

1 day ago | 0

Answered

How to plot Network performance?

You have lost training information, So the only thing left is output vs input. Hope this helps. Thank you for formally ac...

How to plot Network performance?

You have lost training information, So the only thing left is output vs input. Hope this helps. Thank you for formally ac...

4 days ago | 0

Answered

Elman Neural Network (ENN)

size(P_TRAIN) = [ 1296 1728] size(T_TRAIN) = [ 432 1728] Hope this helps. *Thank you for formally accepting my answer* ...

Elman Neural Network (ENN)

size(P_TRAIN) = [ 1296 1728] size(T_TRAIN) = [ 432 1728] Hope this helps. *Thank you for formally accepting my answer* ...

4 days ago | 0

Answered

NARX re-training in closed loop

Using 100 feedback delays makes no sense. Only use feedback delays that are within the correlation length of the function. See...

NARX re-training in closed loop

Using 100 feedback delays makes no sense. Only use feedback delays that are within the correlation length of the function. See...

18 days ago | 1

Answered

Can CNN train separately instead of learning everything at one time?

I try to have the order of inputs as random and uncorrelated as possible. Otherwise the probability of extensive learning/unlear...

Can CNN train separately instead of learning everything at one time?

I try to have the order of inputs as random and uncorrelated as possible. Otherwise the probability of extensive learning/unlear...

19 days ago | 0

Answered

In evaluating a neural net, should NMSE be based only on test subset of data?

For serious work I calulate FOUR values of NMSE: 1.70% Training 2.15% Validation 3.15% Test 4.100% All for 10 (typically...

In evaluating a neural net, should NMSE be based only on test subset of data?

For serious work I calulate FOUR values of NMSE: 1.70% Training 2.15% Validation 3.15% Test 4.100% All for 10 (typically...

1 month ago | 0

| accepted

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....

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....

2 months ago | 0

| accepted

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...

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...

2 months 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 ...

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 ...

2 months ago | 0

| accepted

Answered

transfer function purelin equation based on neural network toolbox

Read the documentation help purelin doc purelin Greg

transfer function purelin equation based on neural network toolbox

Read the documentation help purelin doc purelin Greg

2 months 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...

How to use a sequenceInputLayer with a regressionLayer (neural networks) ?

The answer is obvious: help regressionlayer doc regressionlayer *Thank you for formally accep...

3 months 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...

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...

3 months 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 ...

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 ...

3 months 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...

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...

3 months 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...

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...

3 months 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...

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...

3 months 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...

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...

3 months 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...

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...

3 months 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...

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...

3 months ago | 0

| accepted

Answered

How to plot only few classes for confusion matrix?

Oviously, you have to subsample the original matrices. Greg

How to plot only few classes for confusion matrix?

Oviously, you have to subsample the original matrices. Greg

3 months 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

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

3 months 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 ...

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 ...

3 months 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

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

3 months 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...

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...

3 months ago | 0

| accepted

Answered

Neural Network input and targets have different of samples

Transpose both matrices. Thank you for formally accepting my answer Greg

Neural Network input and targets have different of samples

Transpose both matrices. Thank you for formally accepting my answer Greg

3 months 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 ...

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 ...

3 months 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...

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...

3 months 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

Expressing equation in terms of sin/cos

If you substitute your solutions into LHS and get the RHS, then it is possible. Greg

3 months 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...

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...

3 months 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...

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...

4 months ago | 0