引導式測驗 · 硬 等級
貝葉斯深度學習測驗
探索貝葉斯網路、不確定權重、後驗近似、蒙地卡羅遺失以及保留資料的置信度校準。
問題 1 的 8
貝葉斯神經網路對待權重的方式與標準神經網路有何不同?
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。
- 貝葉斯神經網路對待權重的方式與標準神經網路有何不同?
- 它將每個權重視為機率分佈而不是單一固定值
- 它使用整數權重而不是小數
- 它完全消除了重量
- 它在所有層之間共享一個權重
- What can a Bayesian approach provide that a single fixed-weight network does not directly represent?
- A guarantee that every confidence score is calibrated
- A distribution of predictions that reflects uncertainty about model weights
- A proof that the model will detect every unfamiliar input
- A way to remove all noise from the data
- How can epistemic uncertainty change as relevant, representative training data are added?
- It is guaranteed to reach zero with any added data
- It is identical to label noise
- It may decrease under the chosen model and data assumptions
- It always increases linearly
- 為什麼需要變分推理或 MCMC 等近似方法?
- 因為神經網路沒有損失函數
- 因為GPU不能做加法
- 因為權重必須是整數
- 因為數百萬個權重的精確後驗在計算上是難以處理的
- 蒙特卡羅 dropout 是如何估計不確定性的?
- 在測試時保持 dropout 處於活動狀態,並利用輸出的分佈多次運行網絡
- 關閉輟學並延長訓練時間
- 永久移除所有神經元
- 用隨機噪音代替損失
- 變分推理將後驗估計重新定義為什麼樣的問題?
- 一個排序問題
- 最小化 KL 散度的最佳化問題(透過 ELBO)
- 沒有目標的純隨機抽樣問題
- 硬體設計問題
- How does Bayes’ rule relate a prior, likelihood, and posterior?
- The posterior is the prior plus the likelihood, without normalization
- It is the likelihood divided by the prior without normalization
- The posterior is proportional to the prior times the likelihood
- It is the fixed point estimate of the most likely weight vector
- How does MCMC approximate a Bayesian neural-network posterior?
- By training a single point estimate with dropout disabled
- By ranking model outputs with a fixed rule
- By minimizing an ELBO with a simpler variational family
- By drawing samples through a Markov chain whose stationary distribution targets the posterior