By “technical issues” I think that Jaynes means the measure theory basis of probability, not the everyday applications of statistics.
EDIT: And he’s comparing Kolmogorov’s definition of probability (which starts with a set of possible outcomes and then treats events as subsets) to his own view (IIRC he thinks you can calculate with events directly and need no underlying set). He’s not comparing Frequentism and Bayesianism.
Not directly, no. But you would be hard-pressed to call Kolmogorov a Bayesian, at least when discussing axiomatic probability theory and not complexity/minimum description length, given that he never talked about “the degree of belief”, only about axioms of probability. Yet his results match those of Jaynes (the other way around, really).
Further discussion on Kolmogorov and Bayes is in this paper posted on Luke’s old blog. For example:
For an agent to be perfectly rational, her degrees of belief must obey the axioms of probability theory.
By “technical issues” I think that Jaynes means the measure theory basis of probability, not the everyday applications of statistics.
EDIT: And he’s comparing Kolmogorov’s definition of probability (which starts with a set of possible outcomes and then treats events as subsets) to his own view (IIRC he thinks you can calculate with events directly and need no underlying set). He’s not comparing Frequentism and Bayesianism.
Not directly, no. But you would be hard-pressed to call Kolmogorov a Bayesian, at least when discussing axiomatic probability theory and not complexity/minimum description length, given that he never talked about “the degree of belief”, only about axioms of probability. Yet his results match those of Jaynes (the other way around, really).
Further discussion on Kolmogorov and Bayes is in this paper posted on Luke’s old blog. For example: