Howle L E, Weber P W, Nichols J M
Mechanical Engineering and Materials Science Department, Duke University, 144 Hudson Hall, Durham, NC 27708-0300, United States; BelleQuant Engineering, PLLC, Mebane, NC 27302-9281, United States.
Mechanical Engineering and Materials Science Department, Duke University, 144 Hudson Hall, Durham, NC 27708-0300, United States; BelleQuant Engineering, PLLC, Mebane, NC 27302-9281, United States.
Comput Biol Med. 2017 Mar 1;82:3-11. doi: 10.1016/j.compbiomed.2017.01.006. Epub 2017 Jan 17.
We examine both maximum likelihood and Bayesian approaches for estimating probabilistic decompression sickness model parameters. Maximum likelihood estimation treats parameters as fixed values and determines the best estimate through repeated trials, whereas the Bayesian approach treats parameters as random variables and determines the parameter probability distributions. We would ultimately like to know the probability that a parameter lies in a certain range rather than simply make statements about the repeatability of our estimator. Although both represent powerful methods of inference, for models with complex or multi-peaked likelihoods, maximum likelihood parameter estimates can prove more difficult to interpret than the estimates of the parameter distributions provided by the Bayesian approach. For models of decompression sickness, we show that while these two estimation methods are complementary, the credible intervals generated by the Bayesian approach are more naturally suited to quantifying uncertainty in the model parameters.
我们研究了用于估计概率性减压病模型参数的最大似然法和贝叶斯方法。最大似然估计将参数视为固定值，并通过重复试验确定最佳估计值，而贝叶斯方法将参数视为随机变量并确定参数概率分布。我们最终想知道参数处于某个范围内的概率，而不仅仅是对估计量的可重复性做出陈述。虽然这两种方法都是强大的推理方法，但对于具有复杂或多峰似然性的模型，最大似然参数估计可能比贝叶斯方法提供的参数分布估计更难解释。对于减压病模型，我们表明，虽然这两种估计方法是互补的，但贝叶斯方法生成的可信区间更自然地适用于量化模型参数中的不确定性。
