Bayesian probability theory Redwood Center （贝叶斯概率理论红杉中心）.pdf

Bayesian probability theory Bruno A. Olshausen∗ March 1, 2004 Abstract Bayesian probability theory provides a mathematical framework for peform- ing inference, or reasoning, using probability. The foundations of Bayesian probability theory were laid down some 200 years ago by people such as Bernoulli, Bayes, and Laplace, but it has been held suspect or controversial by mod- ern statisticians. The last few decades though have seen the occurrence of a “Bayesian revolution,” and Bayesian probability theory is now commonly em- ployed (oftentimes with stunning success) in many scientific disciplines, from astrophysics to neuroscience. It is most often used to judge the relative validity of hypotheses in the face of noisy, sparse, or uncertain data, or to adjust the parameters of a specific model. Here we discuss the basics of Bayesian proba- bility theory and show how it has been utilized in models of cortical processing and neural decoding. Bayes’ rule Bayes’ rule really involves nothing more than the manipulation of conditional proba- bilities. Remember that the joint probability of two events, AB , can be expressed as P (AB) = P (A |B)P (B) (1) = P (B |A)P (A) (2) In Bayesian probability theory, one of these “events” is the hypothesis, H , and the other is data, D, and we wish to judge the relative truth of the hypothesis given the data. According to Bayes’ rule, we do this via the relation P (D |H )P (H )