AIAI Agent Dev Labaiagentdev-lab.com
Sign in to save progressSign in
Menu

AIF-C01 · LESSON 1

Fundamentals of AI and ML20% · 3 topics

Learn the basic concepts and terminology of AI, ML and deep learning, practical AI use cases, and the ML development lifecycle.

1.1Basic AI concepts and terminology

1.1 Explain basic AI concepts and terminologies.

Concept

AI (artificial intelligence) is the umbrella term for technologies that perform human intellectual tasks by machine; a subset of it is ML (machine learning: learning patterns from data), and a subset of ML is deep learning (multi-layer neural networks). Learning types are supervised (labeled data: classification, regression), unsupervised (unlabeled: clustering, dimensionality reduction) and reinforcement learning (learning from rewards). Inference is making predictions with a trained model, and there is real-time inference and batch inference.

How it works / steps

  1. Check whether the problem is classification (categories), regression (numbers) or clustering (grouping)
  2. Decide between supervised and unsupervised based on whether labeled data exists
  3. Deep learning is commonly used for complex data such as images, audio and natural language
  4. Understand the data type (structured/unstructured, time series, labeled)

Workplace example

If past transactions are labeled “fraud,” use supervised classification; if there are no labels and you want to find unusual transactions, use unsupervised anomaly detection.

Common confusions and exam tips

Mini quiz

Q1. General Affairs wants to automatically extract the counterparty's name and contract period from the text of contracts. Which AI field is mainly used?

  1. Computer vision
  2. Natural language processing (NLP)
  3. Speech synthesis
  4. Reinforcement learning
Show answer and explanation

Answer: B Natural language processing (NLP)

Understanding the meaning of text and extracting information is natural language processing. For paper contracts, first convert them to text with OCR (such as Textract).

Q2. Which statement about reinforcement learning is incorrect?

  1. An agent acts in an environment and receives rewards as a result
  2. It is sometimes used in fields such as robot control, game playing and optimizing web ad delivery
  3. It learns, through trial and error, actions that increase the reward
  4. Large amounts of data with correct labels must always be prepared before training
Show answer and explanation

Answer: D Large amounts of data with correct labels must always be prepared before training

Reinforcement learning learns from rewards for actions, not from correct labels. Correct labels are required for supervised learning.

Reference links

1.2Practical AI use cases

1.2 Identify practical use cases for AI.

Concept

AI suits finding patterns in large amounts of data, prediction, automation and personalization. On the other hand, AI may not suit processing that needs strict accuracy with clear rules, cases with little data, or cases where unexplainable decisions are not allowed. AWS has purpose-built AI services (Rekognition: images and video, Comprehend: natural language, Textract: documents, Transcribe: speech → text, Polly: text → speech, Translate: translation, Lex: conversation, Personalize: recommendations).

How it works / steps

  1. Clarify the business problem and the value expected from AI (time, accuracy, revenue)
  2. Judge whether rules can solve it or AI is needed
  3. Check whether an off-the-shelf AI service can solve it
  4. If not, build your own model with SageMaker AI

Workplace example

Use Amazon Personalize for product recommendations on an e-commerce site, Comprehend for sentiment analysis of product reviews, and Textract to extract text from ID card images.

Common confusions and exam tips

Mini quiz

Q1. To build an image classification model, you want to efficiently label tens of thousands of images with correct answers. Which service is the most suitable?

  1. Amazon Polly
  2. SageMaker Ground Truth
  3. Amazon Translate
  4. SageMaker Model Registry
Show answer and explanation

Answer: B SageMaker Ground Truth

Ground Truth is a data labeling service with features to manage work by labelers (internal or external) and improve efficiency with automated labeling.

Q2. Which statement about Amazon Comprehend is incorrect?

  1. It can extract key phrases and entities such as person and organization names from text
  2. It can analyze the sentiment of text (positive, negative and so on)
  3. It can build models that classify documents into your own categories
  4. It is a service for detecting faces in images and identifying the people shown
Show answer and explanation

Answer: D It is a service for detecting faces in images and identifying the people shown

Comprehend is a natural language processing service for text. Image and face analysis is Rekognition.

Reference links

1.3The AI/ML development lifecycle

1.3 Describe the AI/ML development lifecycle.

Concept

The ML lifecycle is a repeating cycle: define business goals → frame the problem as ML → collect and prepare data (preprocessing, feature engineering) → train and tune the model → evaluate → deploy → monitor and retrain. MLOps is the practice of making this flow automated and reproducible. SageMaker AI includes Data Wrangler (data preparation), Feature Store, Pipelines, Model Registry, Model Monitor and more.

How it works / steps

  1. Decide business goals and evaluation metrics (accuracy, recall, business KPIs)
  2. Collect data and handle missing values, normalize and create features
  3. Split into training, validation and test data for training and evaluation
  4. Deploy to an endpoint and monitor drift with Model Monitor
  5. Automate retraining with Pipelines

Workplace example

Automatically retrain a demand forecasting model on new data every month, and approve it in Model Registry and deploy only when it meets the evaluation criteria.

Common confusions and exam tips

Mini quiz

Q1. An Accounting analyst wants to build a sales forecasting model from past data without writing code. Which SageMaker AI feature is the most suitable?

  1. SageMaker Canvas
  2. SageMaker Model Monitor
  3. SageMaker Ground Truth
  4. SageMaker Model Registry
Show answer and explanation

Answer: A SageMaker Canvas

SageMaker Canvas is no-code: you can go from data preparation to building a forecasting model and making predictions entirely through a visual interface.

Q2. Which statement about MLOps is incorrect?

  1. Monitor production model performance and retrain with new data as needed
  2. Record experiment conditions and results so the same results can be reproduced
  3. As a rule, never update a model once deployed, even if its accuracy changes
  4. Run everything from data preparation to training and deployment repeatedly through automated steps
Show answer and explanation

Answer: C As a rule, never update a model once deployed, even if its accuracy changes

MLOps is the practice of continuously running experiment management, reproducibility, automation, monitoring and retraining. Models are updated as the data changes.

Reference links

Lessons are written by this site based on public information and do not guarantee exam content. Product specifications change, so check the latest information at the reference links.

Notes (unofficial, original questions, no pass guarantee)