Neural network learning notes

  Two pieces of the same ice , One piece melts into “1” The shape of the , One piece melts into “2” The shape of the , Is the entropy change of these two processes the same ？

  Motion is three-dimensional , And the temperature is one-dimensional . Why can one-dimensional quantities be used to describe three-dimensional thermal motion ？ Because it is assumed that particles move randomly in equilibrium , The probability in all three directions is the same , So exercise can be used 1 Three dimensions to describe . But if the probability of particles moving in the three dimensions is inconsistent , How can we use the temperature of one dimension to measure 3 Three dimensional motion ？

If temperature is a one-dimensional variable , Then no matter what the reference is , The order of temperature of objects is consistent . But if temperature is a multidimensional variable , The temperature order of the object may be variable .

Such as random selection in the woods 5 tree , By the end of 1 Take tree as the origin to count others 4 Tree to the third 1 The distance between trees and sort . And then with the second 2 Tree is the origin of statistics, and the rest 4 Tree to the third 2 The distance between trees and sort . The first 3,4,5 The order of trees is inconsistent in these two sorts .

For example, after the water molecules in the same position of two pieces of ice melt , One movement to （0,2） spot , Another movement to （2,0） spot , Let the temperature be two-dimensional , If the （0,0） Is the origin, then the temperature of these two molecules is the same . Because they arrive （0,0） The same distance . But if you use （1,0） Is the origin, then obviously （0,2） The temperature should be higher , because （0,2） To （1,0） More distance . At this time, their temperature order changed . So if the temperature is multidimensional, change the origin of the temperature , The order of temperature is not unique , Is relative .

If we can use temperature to measure entropy , Suppose entropy is also a multidimensional variable , Then the order of entropy will not be unique .

	8	6	1	9	3	4	2	7	5
0	27905.1	25605.5	23558.5	13346.8	12572.2	11983.1	8358.6	7822.01	5402.95
	1	2	9	7	8	4	0	5	6
3	35671.2	33389.6	19919.6	17331.4	17015.7	15199.5	12572.2	11703.1	8136.27
	2	9	3	0	1	6	7	8	5
4	19984.9	18523.7	15199.5	11983.1	9577.51	9020.96	7572.7	6106.35	5689.27
	9	3	2	1	6	5	0	4	8
7	20211.5	17331.4	12555.6	9568.94	8983.45	8617.16	7822.01	7572.7	7073.43

So suppose entropy is multidimensional . It can explain the inconsistency of sorting on the morphological number axis . And it is consistent with the assumption that the number of iterations is inversely proportional to entropy .

/**/

The function of force and the classification of form

( molecular A, molecular B)---m*n*k---(1,0)(0,1)

The interaction between molecules is a force , But the interaction between molecules also has an external appearance of morphology , So the interaction between molecules is also the interaction between forms .

( The earth , The moon )---m*n*k---(1,0)(0,1)

The interaction between the earth and the moon is also an interaction of forces , Both the earth and the moon have a characteristic external form , Therefore, the interaction between the earth and the moon based on gravity is also an interaction between forms .

(mnist 0 ,mnist 1)---m*n*k---(1,0)(0,1)

Now use network classification mnist Of 0 and 1, This leads to the third form and the interaction between forms , If the first one is based on the morphological interaction of electromagnetic force , The second is the form interaction based on gravity , Is the third kind also an interaction of forces ？

This judgment should depend on the result of classification , If two molecules form a new stable structure, they become a new molecule , This should be an interaction of electromagnetic forces . And if two molecules are infinitely close and eventually compressed into a small black hole , This interaction should be the interaction of gravity .

So for the third case , If the result of classification is 0 and 1 It forms a new and stable form , This may be the interaction of electromagnetic forces , And if the result of classification is 0 and 1 Finally cannot separate , This should be the interaction of gravity , Because only gravity is unidirectional .

So the result of classification expresses the action type of force , Therefore, the action environment of tectonic force can be controlled by controlling the results of classification . So this analogy seems to indicate , The real cause of force is difference , And the reason why this difference has electromagnetic force , Gravity and other different representations , Because there are different external environments . The so-called force is the response of differences to the environment .