Week 2 - Introduction to AWS AI and ML services
Key Topics
Note
This week covers the fundamentals of AI/ML on AWS, including Generative AI, specific AI services, and the SageMaker suite.
1. AI/ML Fundamentals
The concept of using Artificial Intelligence (AI) to solve business problems dates back to the 1970s with technologies like optical character recognition and expert systems.
The Cloud’s Role
The massive storage and processing power available via cloud computing has fueled recent advancements in:
- Machine Learning (ML)
- Deep Learning (DL)
- Generative AI
Challenges to Adoption
Adopting AI technologies comes with hurdles:
- Lack of Explainability: Hard to understand how the model reached a conclusion.
- Indeterminate Outputs: Results can vary.
- Potential for Bias: Models may reflect biases present in training data.
Candidates for AI
| Good Candidates for AI | Better for Traditional Programming |
|---|---|
| Projects with large, high-quality datasets. | Simple, rule-based logic. |
| Complex data relationships. | Formula-driven solutions. |
| Image or video data processing. | Projects requiring full transparency in decision logic. |
2. AWS AI Services
Amazon Q
A family of Generative AI-powered assistants.
- Amazon Q Business: Ingests internal enterprise data to aid in decision-making, streamline tasks, and spark innovation.
Specialized Services
- Amazon Polly (Text-to-Speech): Uses Deep Learning to convert text into lifelike speech.
- Amazon Rekognition (Vision): Uses Deep Learning to detect objects, scenes, text, and faces in images and videos.
- Amazon Comprehend (NLP): Uses Machine Learning to uncover insights and relationships in unstructured text.
3. SageMaker AI & Bedrock
Amazon SageMaker AI
A comprehensive service to build, train, and deploy ML models.
| SageMaker Tool | Purpose |
|---|---|
| Data Wrangler | Prepare data: Create, transform, and select features from raw data. |
| Ground Truth | Label data: Combines automated and human labeling for datasets. |
| Clarify | Bias & Explainability: Detects bias and explains model predictions. |
| Model Monitor | Monitoring: Detects quality deviations in deployed models. |
| Jumpstart | Pre-built solutions: Accelerate development with ready-to-use models. |
| Canvas | No-Code ML: Build and deploy models without writing code. |
| Studio Classic | IDE: Web-based interface for the entire ML workflow. |
Amazon Bedrock
A managed service for building Generative AI applications with Foundation Models (FMs).
Note
Foundation Models (FMs) are large-scale models trained on vast amounts of unlabeled data using self-supervised learning.
- Key Features: Text generation, image generation, summarization, and chatbots.
- PartyRock: An experimental playground to create AI apps without coding.
Application Parameters
When generating text with LLMs in Bedrock, you can tune these parameters:
- Temperature: Controls “creativity” vs. “predictability”.
- Low: Predictable, safe.
- High: Creative, random, unexpected.
- Top P: Controls the pool of words considered.
- Low: Considers only the most likely words (safe).
- High: Considers a broader list including less likely words (diverse).