- Machine Learning: Foundation of AI. Teaches a machine to learn based on data
- Anomaly Detection: Automatic error detection based on anomaly
- Computer Vision: Interpretation of images and video
- Natural Language Processing (NLP): Interpret written or spoken language
- Knowledge mining: Information extraction from unstructured data
Definition: The foundation for most AI solutions
How does it work:
- Machines (computers) learn from data
- Data Scientists use data to train machine learning models to make predicitons based on data.
- Automated ML (AutoML): Create effective ML models with no expertise needed
- Azure ML Designer: A GUI for no-code development of ML models
- Data and Compute: Cloud-based resources for data scientists tu run experiments
- Pipelines: A way to orchestrate tasks like training, validation, and deployment
Definition: Analyzing data over time to identify unusual changes
Anomaly Detector: An Azure service with an API to create anomaly detection solutions
Definition: Area of AI for visual processing based on interpretation of images and video
Models and capabilities
| Task | Description |
|---|---|
| Image Classification | Classify images based on content. For example is this a car or a bike? |
| Object Detection | Classify individual objects and location within an image using a box |
| Semantic Segmentation | Similar to object detection that uses an overlay to color-code distinct objects |
| Image Analysis | Extract information from images including tags for easier cataloging |
| Face detection, analysis, and recognition | Finds human faces in an image. Can be used with facial geometry to recognize individuals |
| Optical Character Recognition (OCR) | Detect and extract text in images, like a road sign or building number |
Azure Services
- Computer Vision: Analyse images and videos to extract descriptions, tags, objects and text
- Custom Vision: Customized image classification with your own images
- Face: Face detection and facial recognition solutions
- Form Recognizer: Information extraction from scanned documents
Definition: Area of AI that understands written and spoken language
Uses:
- Analyze and interpret text in documents
- Interpret spoken language
- Translattion of written and spoken languages
- Interpret commands
Azure Services
- Language: Analyze text or spoken language to build smart applications
- Translator: Translation service for more than 60 languages
- Speech: Recognize and synthesize speech and translate to other languages
- Azure Bot: Conversational AI with the ability to connect to channels like email, Teams, and web chat
Definition: Describe solutions about extracting information from unstructured data to create a searchable one
Azure Service: Azure Cognitive Search, and enterprise solution for building searchable indexes from private or public assets including analyzing images.
Challenges or risks:
- Bias: Trained data might rely heavily on specific race, or geography
- Errors: Mistakes can cause harm (e.g. autonomus vehicles)
- Exposing data: Non-compliant solutions that don't remove Personal Identifiable Information (PII)
- Accessibility: A solution might not work with individuals with disabilities
- Complex systems: Users must trust how solutions are generated (e.g. from what data)
- Liability: Who/what is liable for decisions?
AI development at Microsoft uses 6 principles:
Fairness: AI systems should treat all people fairly
Reliability & Safety: AI systems should work reliably and safely
Privacy & Security: Respect privacy and consider security at all times, even after deployment
Inclusiveness: Empower everyone regardless of ability, gender, and other factors
Transparency: Systems shouold be understandable
Accountability: People should be accountable. Engineers and designers should work with a governance framework