Theory - Introduction Flashcards
What are the precoditions for using Machine Learning?
→ Data is clearly structured
→ Data semantics is well-defined
→ Data is complete, correct, and
not changing over time
→ Problem is well-defined
What are some cases that may lead fully automated solutions to failure?
→ Data Ambiguous and Incomplete
→ Complex Relationship
→ Semantic Gap / Domain- (and World-) Knowledge
→ Limited Accuracy
What is Human Interaction crucial for?
→ Exploration of Data
→ Generation of Hypotheses
→ Interpretation of Results
→ Steering of the Analysis
→ Hypothesis Evaluation
What are the advantages of computers?
→ Data Storage
→ Computing Power
→ Search
What are the advantages of humans?
→ General Knowledge
→ Perception
→ Creativity
What is Visual Analytics?
Tight Integration of Visual and Automatic Data Analysis Methods for Information Exploration and Scalable Decision Support
What are Mixed-Initiative Systems?
Systems utilizing HUMAN-IN-THE-LOOP, AI-IN-THE-LOOP, or a combination of the two
When are Mixed-Initiative Systems useful?
→ Cost-Risk Tradeoffs
→ Contextualization
→ Multi-Objective Optimization
→ Subjective Analysis
→ Personalization
→ Problem Ambiquity
What is Mixed-Initiative?
A flexible interaction strategy in which each agent (human or computer) contributes what it is best suited at the most appropriate time
What is Interactive ML?
Interactive ML (IML) aims to integrate humans
into the process of insight discovery
What is Explainable ML
Explainable AI (XAI) looks to provide human-readable, as well as interpretable explanations of the decisions made by ML models
What are the objectives of I&XAI?
- Understanding of ML Model Decisions and Behavior
- Diagnosis of ML Model Performance and Applicability
- Refinement of ML Models for Given Users, Tasks, and Data
From which areas does I&XAI draw from?
- Machine Learning & Artificial Intelligence
- Human-Computer Interaction
- Information Visualization & Visual Analytics
- Intelligent Interface Design
- Human-Centered Computing
How do we avoid
miscommunication pitfalls
- Adapt the guidance and analysis over time
- Tailor the design of your interaction workflows
- Combine visual and verbal explanation mediums
- Pick the right metaphors for communication
What type of users do you know of?
- Lay user
- Domain expert
- Decision maker
- ML Developer
- ML Expert
What is Human-Centered ML?
- Design with and for Humans in their given Environments
- Use Participatory Approaches
- Contextualize and Personalize
What is Co-Adaptive Analytics?
Users and systems adapting over time to converge to a common understanding and shared analysis process to solve tasks. Through interaction the agents gather information, up-date their
analysis and prioritize tasks
How is Human-Centered ML contacted?
- Collect Qualitative Data about Things that don’t work for a Problem
- Check Requirements and Tasks
- Prototype and Iterate Solutions
How to learn reward models from diverse sources of human feedback?
- Demonstrations
- Rankings
- Comparisons
- Natural language instructions
What are the categories that can be found in Co-Adaptive Analytics?
- User Teaching
- System Learning
- User Learning
- System Teaching
What factors and dependencies can influence the quality of human feedback?
- Type-Dependency
- Task-Dependency
- Progress-Dependency
What is Insight Provenance?
A historical record of the process and rationale by which an insight is derived during a visual analytics task
List 2 tools used in Visualization for Web
- Vega-Lite
- D3.js
