AI Bot
Data Engineering & Deployment for AI Flashcards
Microsoft Azure AI Fundamentals AI-901 Flashcards
Study our Data Engineering & Deployment for AI flashcards for the Microsoft Azure AI Fundamentals AI-901 exam with 43+ flashcards. View as flashcards, a searchable table, or as a fun matching game.
| Front | Back |
| A B testing for models | Compare performance of two models on live traffic segments |
| Autoscaling metrics to monitor | CPU memory request rate latency queue length |
| Batch scoring definition | Process many records periodically for throughput oriented use cases |
| Blob storage purpose | Object storage for unstructured data and large files |
| Blue green deployment definition | Run new version parallel then switch traffic instantaneously |
| Bucketing purpose | Hash based grouping to optimize joins on specific keys |
| Canary deployment definition | Roll out new version to small subset to detect issues early |
| CI CD for ML pipelines | Automate training tests validation and deployment with repeatable steps |
| Columnar storage advantage | Fast reads for analytical queries and better compression |
| Common missing data strategies | Drop rows Impute with mean median or mode Use model based imputation |
| Concept drift definition | Change in relationship between features and target over time |
| Containerization benefits | Consistency across environments isolation and portability |
| Cost optimization strategies | Right size resources spot instances model quantization and caching |
| Data cleaning steps | Remove duplicates Impute missing values Normalize and validate ranges |
| Data drift detection methods | Statistical tests PSI KL divergence and ML based detectors |
| Data governance and lineage | Track data origins transformations and access history |
| Data Lake definition | Centralized repository for raw and processed data at scale |
| Docker image best practice | Small base image multistage builds and explicit dependencies |
| Embeddings for categorical data | Map high cardinal categories to dense continuous vectors |
| Feature crosses purpose | Create interactions between categorical features to capture joint effects |
| Feature engineering transforms | Log scaling binning polynomial features interactions |
| Feature selection methods | Filter Wrapper and Embedded methods |
| Feature store benefits | Centralized feature definitions reuse online and offline consistency |
| Horizontal versus vertical scaling | Horizontal adds instances vertical increases resources per instance |
| Kubernetes role | Orchestrate containers manage scaling and resilience |
| Label encoding caveat | Introduces ordinal relation not suitable for nominal categories |
| Latency versus throughput tradeoff | Lower latency reduces batching and may reduce throughput |
| Logging and tracing importance | Debug issues audit requests and trace end to end flows |
| Major cost drivers for AI deployments | Storage compute data transfer and human operations |
| Model serving endpoints types | REST gRPC and streaming sockets |
| Monitoring metrics for models | Latency throughput error rate accuracy and drift indicators |
| Normalization versus standardization | Normalization scales to 0 1 Standardization centers to mean 0 std 1 |
| One hot encoding when to use | Categorical variables with low cardinality |
| Outlier detection techniques | Z score IQR Isolation Forest and visual inspection |
| Parquet versus CSV | Parquet is columnar compressed schema aware CSV is row oriented text |
| Partitioning benefits | Improves query performance by pruning irrelevant files |
| Real time scoring definition | Serve predictions with low latency per request for immediate responses |
| REST versus gRPC tradeoffs | REST is simple gRPC is high performance binary RPC |
| Security best practices for data storage | Encryption at rest and in transit IAM and network isolation |
| Streaming source examples | Kafka Kinesis Pub Sub |
| Target encoding risk and mitigation | Can leak target information Use cross validation or smoothing |
| Versioning models and data | Keep immutable artifacts tags and provenance for reproducibility |
| When to use SQL databases | Structured data transactional workloads and complex queries |
About the Flashcards
Flashcards for the Microsoft Azure AI Fundamentals exam offer a compact way to review core ML and data engineering terminology. Cards summarize preprocessing steps, data cleaning, imputation, outlier detection, normalization versus standardization, categorical encoding approaches, target-encoding risks, embeddings, feature engineering transforms, selection methods, and feature store basics.
They also cover data storage and formats (data lakes, blob storage, SQL, Parquet vs CSV, columnar layouts, partitioning and bucketing), deployment and serving patterns (batch vs real-time scoring, streaming sources, containerization, Docker, Kubernetes, REST and gRPC, autoscaling, canary and blue-green rollouts), plus monitoring, drift detection, governance, security, versioning, CI/CD, and cost optimization strategies.
Topics covered in this flashcard deck:
- Data cleaning and imputation
- Categorical encoding techniques
- Feature engineering and selection
- Storage formats and partitioning
- Containerization and deployment
- Monitoring, drift, governance