AWS AIF-C01 Responsible AI

Responsible AI 負責任的AI

Bias-variance trade-offs

Variance

. Variance refers to the model’s sensitivity to fluctuations or noise in the training data. 模型對訓練數據中的波動或雜訊的敏感度。

underfitted

underfitting the data because it is not capturing all the features of the data.模型擬合數據不足，因為它沒有捕獲數據的所有特徵

overfitted

capturing noise and is essentially memorizing the data. It won’t perform well on new data. 在捕獲雜訊，本質上是在記憶數據。它對新數據的性能不佳

• 訓練時很強，實際上場很弱，Ben Simmons 型選手

balanced

the bias is low and the variance is low 偏差較低，方差較低。

overcome bias and variance errors

Cross-validation

by training several ML models on subsets of the available input data and evaluating them on the complementary subset of the data
在可用輸入數據的子集上訓練多個 ML 模型，然後在數據的互補子集上評估它們

Increase Data

Regularization

penalizes extreme weight values to help prevent linear models from overfitting

則化是一種懲罰極端權重值的方法，有助於防止線性模型過度擬合訓練數據示例。

Simpler model

overfitting-> easier,
underfitting-> harder

dimension reduction

unsupervised machine learning algorithm that attempts to reduce the dimensionality (number of features)

無監督機器學習演算法，它試圖降低數據集中的維度（特徵數量）

End training early

Transparent and Explainable

Transparency answers the question HOW, and explainability answers the question WHY

透明度有助於瞭解模型如何做出決策, 可解釋性有助於理解模型做出決策的原因。

Explainability frameworks

• SHapley Value Added (SHAP)

• Local Interpretable Model-Agnostic Explanations (LIME),
  help summarize and interpret the decisions made by AI systems 説明總結和解釋 AI 系統做出的決策

Transparent Documentation

Monitored and audited

oversight by humans and automated tools to identify unusual patterns or decisions.人工的定期測試和監督，以及識別異常模式或決策的自動化工具。

human oversight and involvement

counterfactual explanations

user interfaces explanations

Tools for transparency and explainability

AWS AI Service Cards

form of responsible AI documentation that provides customers with a single place to find information on the intended use cases and limitations

負責任的 AI 文件形式，它為客戶提供了一個位置來查找有關預期使用案例和限制、負責任的 AI 設計選擇

SageMaker Model Cards

risk rating of a model, training details and metrics, evaluation results and observations, and additional callouts such as considerations, recommendations, and custom information.

預期用途和風險評級、訓練詳細資訊和指標、評估結果和觀察結果等資訊，以及其他標註，例如注意事項、建議和自定義資訊。

Tools for explainability

SageMaker Clarify SageMaker

help determine if a particular model input has more influence than expected on overall model behavior.

幫助確定特定模型輸入對整體模型行為的影響是否大於預期。

SageMaker Autopilot SageMaker

help ML engineers, product managers, and other internal stakeholders understand model characteristics.
幫助 ML 工程師、產品經理和其他內部利益相關者瞭解模型特徵