9 Questions 24 Answers 0 Followers
Questions related from Chenguang Lu
My studies show that the popular convergence proof of the EM algorithm is not bilievable. An improved algorithm, CM algorithm, might ensure the convergence of mixture models. see...
21 October 2017 8,916 0 View
If NAT(nucleic acid test) shows negative and CT shows positive, which should we believe? Can we calculate their degrees of confirmation to make decision? There are many confirmation measures. Are...
01 January 1970 825 12 View
I found there is a new trend in machine learning, especially in deep learning: researchers use similarity function proportional to P(x|yj)/P(x)=P(yj|x)/P(yj) to construct estimated mutual...
01 January 1970 3,864 1 View
I found the popular convergence proof of the EM algorithm is wrong because Q may and should decrease in some E steps; P(Y|X) from the E-step is also improper Shannon's channel. For details, see my...
01 January 1970 6,953 3 View
Two different viewpoints can be found as follows: Confirmation, causation, and simpson's paradox Confirmation, causation, and simpson's paradox Causal Confirmation Measures: From Simpson’s...
01 January 1970 6,756 3 View
I found the popular convergence proof with EM algorithm for mixture models is incorrect because Q may and should decrease in some cases. For details see From the EM Algorithm to the CM-EM...
01 January 1970 6,240 1 View
My answer is that we can replace distortion constraint with semantic constraint and then use Shannon's information theory to optimize semantic communication. In this way, we can keep existing...
01 January 1970 5,041 0 View
I support Darwin because I found that we can explain birds beauty preferences by the need relationships between birds and their food resources or...
01 January 1970 3,083 0 View
A new explanation is that first needing relationships selected brids tasts, and laterly the female's tasts select the male's plumages. A video and a paper show how birds' colorful plumages reflect...
01 January 1970 8,726 0 View