AI Terms 2 Flashcards
What is a Knowledge Graph?
Network structure representing relationships between entities:
Stores information as interconnected nodes and edges
Enables semantic search and reasoning
Supports relationship discovery
Used in recommendation systems
What is a Graph Database?
Database optimized for storing and querying graph structures:
Native support for relationships
Efficient traversal operations
No need for complex joins
Example: Neo4j
What are ACID Transactions?
Properties ensuring database reliability:
Atomicity: All or nothing execution
Consistency: Valid state transitions
Isolation: Transaction independence
Durability: Permanent changes
What is Vertex AI Matching Engine?
GCP’s vector database service:
Optimizes similarity search
Scales to billions of vectors
Low-latency retrieval
Integrates with Vertex AI
What is Reinforcement Learning?
Learning through environment interaction:
Uses rewards/penalties
Learns optimal actions
Explores vs exploits
Example: Game AI
What is Supervised Learning?
Learning from labeled data:
Input-output pairs
Classification/regression
Requires labeled datasets
Example: Spam detection
What is Unsupervised Learning?
Learning patterns without labels:
Clustering
Dimensionality reduction
Pattern discovery
Example: Customer segmentation
What is Continuous Evaluation?
Ongoing model performance monitoring:
Real-time metrics
Performance degradation detection
Automated testing
Quality assurance
What is Data Drift?
Changes in input data distribution:
Feature value shifts
Input pattern changes
Requires monitoring
May trigger retraining
What is Model Drift?
Degradation of model performance:
Relationship changes
Accuracy decline
Required retraining indicators
Performance monitoring
What are Model Parameters?
Learned variables during training:
Weights and biases
Adjusted automatically
Learned from data
Define model behavior
What is Learning Rate?
Step size for model updates:
Controls convergence speed
Affects training stability
Hyperparameter
Typically 0.001-0.1
What is Batch Size?
Training samples per iteration:
Affects memory usage
Impacts training speed
Trade-off with accuracy
Hyperparameter
What are Batch Workloads?
Processing large data volumes:
Non-real-time
Resource-intensive
Scheduled processing
Bulk operations
What are Real-time Workloads?
Immediate data processing:
Low latency
Stream processing
Immediate responses
Online serving
What is Deterministic Processing?
Fixed outcome for given input:
Predictable results
No randomness
Reproducible
Rule-based
What is Predictive Analysis?
Forecasting future outcomes:
Based on patterns
Probability-based
Uses historical data
What might happen
What is Prescriptive Analysis?
Recommending actions:
Suggests solutions
Optimization-based
What should be done
Action-oriented
What are Cold Start Problems?
Initial recommendation challenges:
New users/items
No historical data
Limited predictions
Requires special handling
What does “Scales Quadratically” mean?
Resource needs grow with square of input:
O(n²) complexity
Exponential resource growth
Performance consideration
Common in attention mechanisms
What is Neural Architecture Search?
Automated network design:
Optimizes model structure
AutoML component
Searches architecture space
Automated optimization
What is Contextualization in AI?
Process of understanding input in its full context:
Considers surrounding information
Improves accuracy
Enhances relevance
Critical for NLP tasks
What is Automated Parameter Adjustment?
Automatic optimization of model parameters:
During training process
Uses gradient descent
Optimizes weights/biases
No manual intervention
What is Manual Hyperparameter Optimization?
Human-guided tuning of model settings:
Trial and error
Expert knowledge
Grid/random search
Performance tracking
