ML Exam Prep: 9 - Misc
ML Exam Prep
Misc
Search Types
1) Lexical search: matches exact words
2) Semantic search: understands concept
3) Contextual search: adapts results using user data or surrounding conversation
Deployment Types
Canary = deploy to alpha sites (small subset of users or servers).
A/B testing = concurrent. some goes to new and rest to production.
Shadow = incoming are mirrored or copied and sent to production and new potential. New potential is logged and analyzed on metrics, but *only* answers from production are returned to end-users.
Blue/Green = All goes to “blue“ (production). Once "green" (new version) is ready, traffic switched over. Has a safe rollback mechanism and min downtime. No parallel testing of new model.
Hyperparameters
Temperature = chaos\creative of response
Top K = is # of tokens. More tokens = more diverse/chaos of responses.
Misc
Categorical model = classification model.
Data Bias & Security
Pre-Training Bias
MetricsClass Imbalance (CI): One category drastically outnumbers others in data.
Diff in Proportions of Labels (DPL): Measures imbalanced positive outcomes across facets.
SageMaker Clarify: Primary AWS tool to detect and mitigate ML bias.
Diff in Proportions of Labels (DPL): Measures imbalanced positive outcomes across facets.
SageMaker Clarify: Primary AWS tool to detect and mitigate ML bias.
Handling Bias
Resampling: Oversample minority classes or undersample majority classes.Synthetic Data: Use SMOTE to generate fake minority class examples.
Augmentation: Crop, flip, or rotate images to expand training variety.
Synthetic Data: Use SMOTE to generate fake minority class examples.
Augmentation: Crop, flip, or rotate images to expand training variety.
Augmentation: Crop, flip, or rotate images to expand training variety.
Security & Compliance
Encryption: Use S3 server side encryption with KMS for data at rest and TLS for data in transit.Anonymization: Mask or tokenize PII (Personally Identifiable Information).
Compliance: Ensure data storage locations respect PHI (HIPAA) and data residency laws.External-account: For external company access to S3 bucket in the company's AWS account: 1) Use an identity-based policy to attach to IAM users that grants or denies access to the S3 bucket.
2) Create new IAM role with a trust policy in the company's account for the external account with permissions and specify the external company's account ID.
Compliance: Ensure data storage locations respect PHI (HIPAA) and data residency laws.
External-account: For external company access to S3 bucket in the company's AWS account:
1) Use an identity-based policy to attach to IAM users that grants or denies access to the S3 bucket.
2) Create new IAM role with a trust policy in the company's account for the external account with permissions and specify the external company's account ID.
Comments
Post a Comment