machine learning - Covers function counting theorem for perceptron -
Looking at the P point in the normal position in the new dimensional space. Probably 2 ^ P possible scorpions with two labels, how many of them are linearly different? Answer to this question by the coir theorem: P>
However, it seems that I can not understand the theorem properly. If I take P = 4, N = 2 theorem yields 8 possibilities If I had 4 digits in 2 dimensional space then I could:
1) All the remaining 4 points 1 or 1 to the rest Separate giving 2 possibilities with labels
2) Separate remaining 3 to 1 points where single point labels (+1, -1) can give 2 more possibilities
3) 2 points apart from the remaining 2 digits, giving me two more possibilities with the same logic as before.
If I now want to separate from 1 to 3 points, then I treat the case which was previously treated in part 1). So it seems that something wrong happened to me Can anyone tell me that what is the meaning of this theorem?
It breaks because in R ^ 2 you are 4 digits in normal position.
Comments
Post a Comment