引導式測驗 · 中等 等級
MAP 估計與先驗測驗
將似然性和先驗與後驗模式連結起來,識別 L1 和 L2 先驗對應關係,並將 MAP 與完全後驗推理區分開來。
問題 1 的 8
Which quantity does MAP maximize?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- Which quantity does MAP maximize?
- The posterior density given the observed data
- The likelihood alone, regardless of the prior
- The prior density alone, regardless of observed data
- The marginal likelihood integrated over all parameter values
- What does MLE maximize compared with MAP?
- The posterior including a prior
- The likelihood
- The prior density only
- 後間隔寬度
- Which prior commonly yields an L2-style coefficient penalty in MAP?
- A Laplace prior, whose log density gives an absolute-value penalty
- A uniform prior, which is constant within its support
- A zero-centered Gaussian prior, whose negative log density gives a squared penalty
- A point mass at zero, which fixes every coefficient to zero
- Which prior commonly yields an L1-style penalty?
- A Gaussian prior, which gives a squared-magnitude penalty
- A uniform prior, which gives no penalty within its support
- A point mass at zero, which fixes the parameter
- A Laplace prior, whose negative log density gives an absolute-value penalty
- Why can MAP differ from MLE with limited data?
- MAP discards the likelihood when data are limited
- The prior can pull the estimate toward values it favors
- MLE always returns a zero estimate with small samples
- MAP selects the single observation with the largest value
- What does a MAP point estimate provide by itself?
- A point estimate that also specifies a credible interval
- The full posterior distribution over all parameter values
- The posterior mode, without a full uncertainty summary
- A guarantee that the assumed prior is correct
- Why can prior sensitivity analysis be informative?
- It guarantees that posterior and prior distributions are identical
- It proves that a likelihood is misspecified whenever the MAP changes
- It removes the need to inspect observed data
- It shows whether conclusions change under plausible prior choices
- Why may a posterior mode change under reparameterization?
- A density changes with the coordinate system, so its mode need not map invariantly
- The likelihood stops depending on observations after transformation
- A posterior mode is invariant by definition for every nonlinear map
- Reparameterization converts a continuous parameter into a class label