ai900 Artificial Intelligence workloads and considerations part 2

 

1. Identify features of common AI workloads

AI workloads = tasks AI systems perform.

  • Computer Vision – analyzing images/videos.

  • NLP – understanding and generating human language.

  • Document Processing – extracting info from forms, PDFs, etc.

  • Conversational AI – chatbots, virtual agents.

  • Generative AI – creating new text, images, code, or media.

Keypoint: Workload = category of AI use.
Hint: Match workload to real-world example.

2. Identify computer vision workloads

Computer vision = AI that interprets visual input.

  • Image Classification – identify object type (cat vs dog).

  • Object Detection – locate objects in an image.

  • Semantic Segmentation – label pixels (road, car, tree).

  • Face Recognition – identify people.

  • OCR (Optical Character Recognition) – read text from images.

Keypoint: Anything with images/videos.
Hint: If it “sees,” it’s computer vision.

3. Identify natural language processing (NLP) workloads

NLP = AI that understands and uses human language.

  • Text Classification – spam vs non-spam email.

  • Sentiment Analysis – detect positive/negative review.

  • Named Entity Recognition (NER) – find names, places, dates.

  • Language Translation – English → French.

  • Speech Recognition & Synthesis – voice to text, text to speech.

Keypoint: Anything with text or speech.
Hint: If it “reads, writes, or talks,” it’s NLP.

4. Identify document processing workloads

Document processing = AI extracts and organizes data from documents.

  • Form Recognizer – get key-value pairs from invoices, receipts.

  • Classification – sort docs (invoice vs contract).

  • OCR for Docs – scan PDFs/images to text.

  • Data Extraction – pull out totals, dates, names.

Keypoint: Focused on structured/unstructured documents.
Hint: Think “AI reads and fills forms.”

5. Identify features of generative AI workloads

Generative AI = creates new content.

  • Text Generation – essays, code, chat.

  • Image/Video Generation – AI art, avatars.

  • Audio/Music Generation – voices, soundtracks.

  • Code Generation – write/assist with programming.

Keypoint: Output is new content, not just analysis.
Hint: If AI “creates,” it’s generative AI.


Identify guiding principles for responsible AI

1. Guiding principles for responsible AI

  • Fairness – AI should not discriminate.

  • Reliability & Safety – AI should work consistently and safely.

  • Privacy & Security – protect user data and keep it safe.

  • Inclusiveness – AI should be accessible to all people.

  • Transparency – AI decisions should be explainable.

  • Accountability – humans are responsible for AI outcomes.

Hint: Remember 6 pillars.

2. Fairness in an AI solution

  • AI should treat all users equally.

  • Avoid bias (gender, race, age, language).

  • Use diverse training data to reduce discrimination.

Example: Loan approval system must not favor one gender.

Hint: Think “no bias, equal treatment.”

3. Reliability and Safety in an AI solution

  • AI must work as expected in all conditions.

  • Must handle errors and unexpected inputs.

  • Should not cause harm.

  • Regular testing and monitoring needed.

Example: Self-driving car must stop safely in all conditions.

Hint: Think “works every time, safe to use.”

4. Privacy and Security in an AI solution

  • Protect personal data (don’t expose user info).

  • Follow laws (GDPR, etc.).

  • Secure AI models against misuse or attacks.

Example: Voice assistant must not leak recorded conversations.

Hint: Think “protect data, protect system.”

5. Inclusiveness in an AI solution

  • AI should work for people of all abilities, languages, cultures.

  • Provide accessibility features (speech-to-text, screen readers).

Example: Translator app works for multiple languages, including regional.

Hint: Think “AI for everyone.”

6. Transparency in an AI solution

  • Users should know how AI makes decisions.

  • Explainable AI = clear reasoning behind outputs.

  • Avoid “black box” models.

Example: Credit scoring AI shows why loan was denied.

Hint: Think “clear and explainable.”

7. Accountability in an AI solution

  • Final responsibility is with humans, not AI.

  • Organizations must monitor and correct AI misuse.

  • Define who is responsible for decisions.

Example: Hospital, not AI system, is responsible for medical diagnosis errors.

Hint: Think “humans answer for AI.”

Describe fundamental principles of machine learning on Azure (15-20%)

1. Common Machine Learning Techniques

  • Regression – predict numeric values (price, temperature).

  • Classification – assign labels (spam/not spam).

  • Clustering – group similar data without labels (customer segments).

Hint: Regression = numbers, Classification = categories, Clustering = groups.

2. Regression Machine Learning Scenarios

  • Predict continuous values.

  • Example: Predicting house price, sales forecast.

Hint: Output = number.

3. Classification Machine Learning Scenarios

  • Predict categories.

  • Binary (Yes/No) or Multi-class (Cat/Dog/Bird).

  • Example: Spam detection, disease diagnosis.

Hint: Output = label.

4. Clustering Machine Learning Scenarios

  • Unsupervised learning.

  • Group data by similarity without pre-defined labels.

  • Example: Customer segmentation, market analysis.

Hint: No labels, AI finds groups.

5. Features of Deep Learning Techniques

  • Uses neural networks with many layers.

  • Good for complex tasks: images, speech, natural language.

  • Needs large datasets + high compute (GPUs).

Hint: Think “big data + many layers.”

6. Features of the Transformer Architecture

  • Deep learning model specialized for NLP.

  • Uses attention mechanism to focus on important words.

  • Basis for models like GPT, BERT.

  • Handles sequential data better than RNNs.

Hint: Think “language + attention.”

7. Core Machine Learning Concepts

Features and Labels in a Dataset

  • Features = input variables (height, weight).

  • Label = target output (disease: yes/no).

Hint: Features → predict Label.

Training and Validation Datasets

  • Training data – teaches model.

  • Validation data – tests accuracy on unseen data.

  • Prevents overfitting.

Hint: Train = learn, Validate = check.

8. Azure Machine Learning Capabilities

  • End-to-end ML service.

  • Build, train, deploy, and manage models.

  • Provides studio, SDK, CLI.

  • Supports automated ML, pipelines, and MLOps.

Hint: One-stop ML platform.

9. Capabilities of Automated Machine Learning (AutoML)

  • Automatically selects algorithm, features, hyperparameters.

  • Reduces manual effort.

  • Useful for beginners.

Hint: “AI builds AI model.”

10. Data and Compute Services for Data Science & ML

  • Data: Azure Data Lake, Blob Storage, SQL Database.

  • Compute:

    • CPU/GPU clusters for training.

    • Azure Machine Learning Compute.

    • Azure Databricks for big data.

Hint: Data = storage, Compute = training power.

11. Model Management and Deployment in Azure ML

  • Model Registry – store versions.

  • Deployment – as web services (REST API).

  • Monitoring – track performance, drift, retrain if needed.

Hint: Train → Store → Deploy → Monitor.


Describe features of computer vision workloads on Azure (15–20%)

1. Common Types of Computer Vision Solutions

  • Image Classification – identify what’s in an image.

  • Object Detection – find and locate objects.

  • OCR (Optical Character Recognition) – extract text from images/docs.

  • Facial Detection & Analysis – detect faces, emotions, age, etc.

Hint: If it “sees” or “reads images,” it’s computer vision.

2. Features of Image Classification Solutions

  • Assigns an image to a single/multiple categories.

  • Example: Cat vs Dog, hotdog vs not-hotdog.

  • Multi-class = one label, Multi-label = multiple labels.

Hint: Whole image = one/more categories.

3. Features of Object Detection Solutions

  • Identifies objects inside an image and their locations.

  • Uses bounding boxes.

  • Example: Detect cars, pedestrians in traffic camera.

Hint: Finds what and where.

4. Features of Optical Character Recognition (OCR)

  • Extracts text from images, PDFs, handwritten notes.

  • Example: Invoice scanning, number plate recognition.

  • Outputs machine-readable text.

Hint: AI “reads text in pictures.”

5. Features of Facial Detection & Facial Analysis

  • Facial Detection – locate faces in images.

  • Facial Analysis – estimate age, gender, emotion, landmarks.

  • Example: Security check, emotion analysis in apps.

Hint: Detects and describes human faces.

6. Azure Tools and Services for Computer Vision Tasks

  • Azure AI Vision – OCR, image classification, object detection.

  • Azure AI Face – face detection, recognition, analysis.

  • Custom Vision – build/train your own image classifiers.

Hint: Vision = general, Face = people, Custom Vision = train your own.

7. Capabilities of Azure AI Vision Service

  • OCR (printed & handwritten text).

  • Image analysis (tags, captions, categories).

  • Object detection.

  • Spatial analysis (counting people in space).

Hint: All-in-one vision service.

8. Capabilities of Azure AI Face Detection Service

  • Detect faces in images/videos.

  • Identify/verify individuals.

  • Analyze emotions, age, gender, head pose, landmarks.

  • Supports face recognition for authentication.

Hint: Specialized for people’s faces.

Describe features of Natural Language Processing (NLP) workloads on Azure (15–20%)


1. Features of Common NLP Workload Scenarios

  • Key Phrase Extraction – find important words/phrases.

  • Entity Recognition – detect names, places, dates, etc.

  • Sentiment Analysis – positive/negative/neutral tone.

  • Language Modeling – predict/generate text.

  • Speech Recognition & Synthesis – convert speech ↔ text.

  • Translation – convert one language to another.

Hint: NLP = AI that reads, writes, listens, talks.

2. Key Phrase Extraction

  • Finds the main topics/phrases in text.

  • Example: “Payment delayed due to system error” → Key phrases: payment, system error.

Use: Summarization, search indexing.
Hint: Extracts keywords.

3. Entity Recognition

  • Identifies specific items in text.

  • Example: “John works at Microsoft in Paris” → John = person, Microsoft = organization, Paris = location.

Use: Info extraction, knowledge graphs.
Hint: Finds names, places, dates.

4. Sentiment Analysis

  • Detects tone of text (positive, negative, neutral).

  • Example: “The product is amazing” → Positive.

Use: Customer feedback, social media monitoring.
Hint: Finds feelings in text.

5. Language Modeling

  • Predicts or generates text.

  • Powers chatbots, auto-complete, generative AI.

  • Example: “I am going to the …” → predicts “store.”

Use: Conversational AI, text generation.
Hint: Think next word prediction.

6. Speech Recognition & Synthesis

  • Speech Recognition – speech → text.

  • Speech Synthesis (Text-to-Speech) – text → spoken output.

  • Example: Virtual assistants, dictation tools.

Use: Voice commands, accessibility.
Hint: Converts voice ↔ text.

7. Translation

  • Converts text/speech from one language to another.

  • Example: English → Spanish.

Use: Global communication, multilingual apps.
Hint: Language conversion.

8. Azure Tools & Services for NLP

  • Azure AI Language – key phrase extraction, entity recognition, sentiment, translation, summarization.

  • Azure AI Speech – speech recognition, synthesis, translation.

Hint: Language = text focus, Speech = voice focus.

9. Capabilities of Azure AI Language Service

  • Key phrase extraction.

  • Entity recognition.

  • Sentiment analysis.

  • Summarization, question answering.

  • Translation (text).

Hint: Full text analytics suite.

10. Capabilities of Azure AI Speech Service

  • Speech-to-text (recognition).

  • Text-to-speech (synthesis).

  • Real-time translation (speech-to-speech).

  • Custom voice models.

Hint: Voice-focused AI.

Describe features of generative AI workloads on Azure (20–25%)

1. Features of Generative AI Solutions

  • Create new content (text, images, audio, video, code).

  • Works with unstructured data.

  • Interactive and adaptive (chatbots, content generation).

Hint: Generative = AI creates.

2. Features of Generative AI Models

  • Pre-trained on large datasets.

  • Use deep learning + transformers.

  • Can be fine-tuned for specific tasks.

  • Examples: GPT (text), DALL·E (images), Codex (code).

Hint: Big models, flexible outputs.

3. Common Scenarios for Generative AI

  • Text – chatbots, summarization, content writing.

  • Code – AI-assisted programming.

  • Images/Video – AI art, design, synthetic media.

  • Audio – voice generation, music.

  • Data Simulation – synthetic data for training models.

Hint: Any case where new content is made.

4. Responsible AI Considerations for Generative AI

  • Fairness – avoid bias in generated content.

  • Reliability/Safety – prevent harmful/misleading outputs.

  • Privacy/Security – protect user data.

  • Inclusiveness – accessible to all.

  • Transparency – disclose AI-generated content.

  • Accountability – humans remain responsible.

Hint: Apply Microsoft’s 6 AI principles.

5. Generative AI Services & Capabilities in Microsoft Azure

  • Azure AI Foundry – build, manage, deploy generative AI apps.

  • Azure OpenAI Service – access GPT, DALL·E, Codex via APIs.

  • Model Catalog – browse, compare, and deploy AI models.

Hint: Foundry = workspace, OpenAI = models, Catalog = store.

6. Features & Capabilities of Azure AI Foundry

  • Central hub for AI development.

  • Tools for prompt engineering, testing, deployment.

  • Integration with other Azure services (Data, Compute).

  • Supports responsible AI monitoring.

Hint: End-to-end AI lifecycle.

7. Features & Capabilities of Azure OpenAI Service

  • Provides pre-trained generative models (GPT, Codex, DALL·E).

  • Use via API or SDK.

  • Supports chat completion, embeddings, image generation.

  • Enterprise-grade security & compliance.

Hint: Access OpenAI models securely in Azure.

8. Features & Capabilities of Azure AI Foundry Model Catalog

  • Repository of ready-to-use AI models.

  • Includes open-source + commercial models.

  • Allows search, compare, evaluate, and deploy.

  • Covers domains: NLP, vision, speech, generative AI.

Hint: Like an app store for AI models.
































































prev