How Deep Neural Networks Work

This is another video from Brandon Rohrer. I add another one two posts ago called "What do Neural Networks learn?  ". I add to these posts the tag "Artificial Neural Networks" to come back to see them whenever needed.

ANN are becoming quite popular and there is more a more interest to see how they work and how to apply them to the NIR spectra.

Meanwhile we try to understand as much as possible what can be considered as a black box. Thanks to Bandon for these great tutorial videos.

See an application of NIR technology in:
An application of ANN in Near Infrared (Protein in wheat with Infratec)
 

Box plot spectra

I have been working this day quite a lot with the concept of good product, and the spectrum with boxplots is a niece example to detect samples which can be contaminated or not to be good product.

 

Always in the case that the good product could be the average spectrum of N samples considered or tested that are good, we can define with all the good samples a boxplot spectra, and over-plot over it new samples and see if they are out of the limits at certain wavelengths, so this can be a clue for a contamination, a confusion in the mixture with the percentages or the components of the mixture.

 

 

What do neural networks learn?

See an application of NIR technology in:
An application of ANN in Near Infrared (Protein in wheat with Infratec)

Using RMS statistic in discriminant analysis (.dc4)

In the case we want to check if a certain spectrum belongs to a certain product we can create an algorithm with PCA in such a way that this algorithm try to reconstruct the unknown spectrum with the scores of this unknown spectrum on the PCA space of the product, and the loadings of the product. So we have the reconstructed spectrum of the unknown and the original spectrum of the unknown.

 

If we subtract one from the another we get the Residual spectrum which is really informative. We can calculate the RMS value of this spectrum to see if the unknown spectrum is really well reconstructed so the RMS values is small (RMS is used as statistic to check the noise in the diagnostics of the instrument).

 

Find the right cutoff to check if the sample is well reconstructed depends of the type of sample and sample presentation.

 

Win ISI multiply the RMS by 1000, so the default value for this cutoff which is 100 in reality is 0.1, anyway a smaller or higher value can be used depending of the application.

 

This type of discrimination is known as RMS-X residual in Win ISI 4 and create ".dc4" models.

 

We see in next posts other ways to use this RMS residual.

Correcting skewness with Box-Cox

We can use with Caret the function BoxCoxTrans to correct the skewness. With this function we get the lambda value to apply to the Box-Cox formula, and get the correction. In the case of lambda = 0 the Box-Cox transformation is equal to log(x), if lambda = 1 there are not skewness so not transformation is needed, if equals 2 the square transformation is needed and several math functions can be applied depending of the lambda value.

In the case of the previous post (correcting skewness with logs)if we use the Caret function "BoxCoxTrans", we get this result:

> VarIntenCh3_Trans
Box-Cox Transformation
1009 data points used to estimate Lambda
Input data summary:
  Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
0.8693  37.0600  68.1300 101.7000 125.0000 757.0000
Largest/Smallest: 871
Sample Skewness: 2.39
Estimated Lambda: 0.1
With fudge factor, Lambda = 0 will be used for transformations

So, if we apply this transformation, we will get the same skewness value and histogram than when applying logs.