マイ用語
用語のポップアップで「お気に入り」「苦手」にした用語の一覧です。
お気に入り
- A
- Accessibility
- Accuracy
- Adaptability
- Agent-to-Agent (A2A)
- AgentCore Identity
- AgentCore Policy
- Agentic AI
- AI agent
- AI assistant
- Algorithm
- Amazon Bedrock
- Amazon Bedrock AgentCore
- Amazon Bedrock Guardrails
- Amazon Bedrock Guardrails
- Amazon Bedrock model evaluation
- Amazon Bedrock Prompt Management
- Amazon CloudWatch
- Amazon Comprehend
- Amazon Inspector
- Amazon Lex
- Amazon Macie
- Amazon Nova
- Amazon Personalize
- Amazon Polly
- Amazon Quick
- Amazon Rekognition
- Amazon SageMaker AI
- Amazon SageMaker Clarify
- Amazon SageMaker Model Cards
- Amazon Textract
- Amazon Transcribe
- Amazon Translate
- Anonymization
- Application security
- ApplyGuardrail API
- Artificial Intelligence
- Asynchronous inference
- Audit trails and logging
- Auditability
- Automated reasoning checks
- Automatic evaluation
- Automation
- Autonomy
- Availability
- Average revenue per user (ARPU)
- AWS AI Service Cards
- AWS Artifact
- AWS CloudTrail
- AWS Config
- AWS PrivateLink
- AWS's eight core dimensions
- AWS Trusted Advisor
- AWS Well-Architected Tool
- B testing
- Balanced dataset
- Baseline
- Batch inference
- Benchmark dataset
- BERTScore
- Bias
- Bias drift
- Biased model outputs
- Bilingual Evaluation Understudy
- Binary
- Black box
- Business goal
- Capabilities
- Carbon footprint
- Chain-of-thought (CoT)
- Chunking
- Class imbalance
- Classic / traditional ML
- Classification
- Clustering
- Code generation
- Cold start
- Compare variants
- Compliance
- component-level evaluation
- Computer Vision
- Confidence scoring
- Constraints
- Content filters
- Content generation
- Context
- Context compaction
- Context engineering
- Context relevance
- Context window
- Contextual grounding check
- Continued pre-training
- Conversational ability
- Conversion rate
- Cost
- Cost-benefit analysis
- Cost-effectiveness
- Cost per interaction
- Cost per user
- Cross-domain performance
- Cross-Region inference
- Curated data sources
- Custom-trained model
- Customer feedback
- Customer lifetime value
- Customer service agent
- Customization
- Data access control
- Data catalog
- Data classification
- Data collection
- Data curation
- Data integrity
- Data leak prevention
- Data lifecycle
- Data lineage
- Data preprocessing
- Data provenance
- Data quality assessment
- Data residency
- Data selection
- Decision support
- Deep Learning
- Demographic groups
- Denied topics
- Deployment
- Deterministic outcome
- Diffusion model
- Discovery
- Discriminative
- Distillation
- Diversity
- Domain adaptation
- Draft
- Efficiency
- Embedding
- Embedding dimensions
- Encryption
- Encryption at rest / in transit
- End-to-end
- End-user risk
- Endpoint
- Energy use
- Environmental considerations
- Escalation rate
- Evaluation
- Experimentation
- Explainability
- Exploratory Data Analysis
- Exposure
- F1 score
- Fairness
- Faithfulness / Groundedness
- Feature engineering
- Feedback
- Few-shot
- Fine-tuning
- Fit
- Foundation model (FM)
- Fraud detection
- General
- Generative
- Generative AI
- Generative AI Security Scoping Matrix
- Governance
- Governance framework
- Graph
- Graph database
- Grounding
- Guardrails
- Hallucination
- Hallucination detection
- Hijacking
- Human audit
- Human-centered design
- Human-in-the-loop
- Human-in-the-loop evaluation
- Human review
- Hyperparameter tuning
- IAM roles, policies, permissions
- Image data
- Image / video / audio generation
- Improved response quality
- In-context learning (ICL)
- Inaccuracy
- Inclusivity
- Inference
- Infrastructure protection
- Input data
- Input/output length
- Input tokens
- Instruction tuning
- instructions
- Insufficient data
- Interpretability
- Interpretable model
- IP infringement claims
- Jailbreaking
- Kiro
- Knowledge base
- Label
- Label quality analysis
- Labeled
- Labeled data
- Labeling
- Latency
- Least privilege
- License
- LLM
- LLM-as-a-judge
- Logging
- Loss of customer trust
- Low barrier to entry
- Machine Learning
- Managed API service
- Masking
- Maximum output length
- Meeting business goals
- Memory
- Memory management
- MLOps
- Modality
- Model
- Model complexity
- Model Context Protocol
- Model distillation
- Model monitoring
- Model selection
- Model size
- Model training
- Model type
- Monitoring
- Multi-agent system
- Multiclass
- Multilingual
- Multimodal model
- Negative prompt
- Neural Network
- NLP
- Non-determinism
- Observability
- On-demand
- Open-source models, data, licenses
- Open-source pretrained model
- Operational constraints
- Operational overhead
- Orchestration
- Output filtering and validation
- Output format
- Output tokens
- Output validation
- Overfitting
- Performance
- Performance requirements
- Personalization
- pgvector
- Poisoning
- Policy
- Post-hoc explanation
- Pre-training
- Precision
- Prediction
- Privacy-enhancing technologies
- Production readiness
- Productivity
- Prompt
- Prompt attack protection
- Prompt builder
- Prompt caching
- Prompt-completion pair
- Prompt engineering
- Prompt injection
- Prompt optimization
- Prompt template
- Prompt version
- Proprietary model
- Provisioned throughput
- RAG grounding
- Real-time inference
- Recall
- Recall-Oriented Understudy for Gisting Evaluation
- Recommendation
- Recommendation system
- Redundancy
- Reference text
- Regional coverage
- Regression
- Regulation
- Reinforcement learning
- Repeatable process
- Representativeness
- Responsibility
- Responsiveness
- results
- Retention
- Retraining
- Retrieval-Augmented Generation (RAG)
- Retrieval quality
- Return on Investment
- Reuse
- Review cycle
- Review strategy
- Reward
- Right-sizing
- RLHF
- Robustness
- Rule-based system
- Safety
- SageMaker JumpStart
- SageMaker Model Cards
- Scalability
- Scalable system
- Search
- Security
- Self-hosted API
- Self-supervised
- Semi-
- Sensitive information filters
- Serverless inference
- Shared responsibility model
- Similarity search
- Simplicity
- Single-shot (one-shot)
- Size
- Source citation
- specific
- Specificity and concision
- Speech recognition
- Speed to market
- Stop sequence
- Strands Agents
- Structured
- Subgroup analysis
- Summarization
- Supervised learning
- Supervisor pattern
- Sustainability
- Swarm
- System prompt
- Tabular data
- Task completion
- Task completion rate
- Task engineering
- Team training requirements
- Technical debt
- Temperature
- Text data
- Threat detection
- Time series data
- Token
- Token-based pricing
- Tool
- Tool-call accuracy
- Tool descriptions
- Tool use
- Top K
- Top P
- Toxicity
- Trade-off
- Traditional ML model
- Training
- Transfer learning
- Transformer-based LLM
- Translation
- Transparency
- Transparency of AI decisions
- Transparency standards
- Trustworthiness
- Truthfulness
- Underfitting
- Unlabeled data
- Unstructured
- Unsupervised learning
- User engagement
- User feedback mechanism
- User satisfaction (CSAT)
- Using multiple comments
- Variable
- Variance
- Vector
- Vector database
- Veracity
- Vulnerability management
- Weights
- Word filters
- Workflow
- Workflow orchestration
- Zero-shot
まだありません。項目ページで用語をクリックし、「お気に入り」か「苦手」を選ぶとここに表示されます。
苦手
- A
- Accessibility
- Accuracy
- Adaptability
- Agent-to-Agent (A2A)
- AgentCore Identity
- AgentCore Policy
- Agentic AI
- AI agent
- AI assistant
- Algorithm
- Amazon Bedrock
- Amazon Bedrock AgentCore
- Amazon Bedrock Guardrails
- Amazon Bedrock Guardrails
- Amazon Bedrock model evaluation
- Amazon Bedrock Prompt Management
- Amazon CloudWatch
- Amazon Comprehend
- Amazon Inspector
- Amazon Lex
- Amazon Macie
- Amazon Nova
- Amazon Personalize
- Amazon Polly
- Amazon Quick
- Amazon Rekognition
- Amazon SageMaker AI
- Amazon SageMaker Clarify
- Amazon SageMaker Model Cards
- Amazon Textract
- Amazon Transcribe
- Amazon Translate
- Anonymization
- Application security
- ApplyGuardrail API
- Artificial Intelligence
- Asynchronous inference
- Audit trails and logging
- Auditability
- Automated reasoning checks
- Automatic evaluation
- Automation
- Autonomy
- Availability
- Average revenue per user (ARPU)
- AWS AI Service Cards
- AWS Artifact
- AWS CloudTrail
- AWS Config
- AWS PrivateLink
- AWS's eight core dimensions
- AWS Trusted Advisor
- AWS Well-Architected Tool
- B testing
- Balanced dataset
- Baseline
- Batch inference
- Benchmark dataset
- BERTScore
- Bias
- Bias drift
- Biased model outputs
- Bilingual Evaluation Understudy
- Binary
- Black box
- Business goal
- Capabilities
- Carbon footprint
- Chain-of-thought (CoT)
- Chunking
- Class imbalance
- Classic / traditional ML
- Classification
- Clustering
- Code generation
- Cold start
- Compare variants
- Compliance
- component-level evaluation
- Computer Vision
- Confidence scoring
- Constraints
- Content filters
- Content generation
- Context
- Context compaction
- Context engineering
- Context relevance
- Context window
- Contextual grounding check
- Continued pre-training
- Conversational ability
- Conversion rate
- Cost
- Cost-benefit analysis
- Cost-effectiveness
- Cost per interaction
- Cost per user
- Cross-domain performance
- Cross-Region inference
- Curated data sources
- Custom-trained model
- Customer feedback
- Customer lifetime value
- Customer service agent
- Customization
- Data access control
- Data catalog
- Data classification
- Data collection
- Data curation
- Data integrity
- Data leak prevention
- Data lifecycle
- Data lineage
- Data preprocessing
- Data provenance
- Data quality assessment
- Data residency
- Data selection
- Decision support
- Deep Learning
- Demographic groups
- Denied topics
- Deployment
- Deterministic outcome
- Diffusion model
- Discovery
- Discriminative
- Distillation
- Diversity
- Domain adaptation
- Draft
- Efficiency
- Embedding
- Embedding dimensions
- Encryption
- Encryption at rest / in transit
- End-to-end
- End-user risk
- Endpoint
- Energy use
- Environmental considerations
- Escalation rate
- Evaluation
- Experimentation
- Explainability
- Exploratory Data Analysis
- Exposure
- F1 score
- Fairness
- Faithfulness / Groundedness
- Feature engineering
- Feedback
- Few-shot
- Fine-tuning
- Fit
- Foundation model (FM)
- Fraud detection
- General
- Generative
- Generative AI
- Generative AI Security Scoping Matrix
- Governance
- Governance framework
- Graph
- Graph database
- Grounding
- Guardrails
- Hallucination
- Hallucination detection
- Hijacking
- Human audit
- Human-centered design
- Human-in-the-loop
- Human-in-the-loop evaluation
- Human review
- Hyperparameter tuning
- IAM roles, policies, permissions
- Image data
- Image / video / audio generation
- Improved response quality
- In-context learning (ICL)
- Inaccuracy
- Inclusivity
- Inference
- Infrastructure protection
- Input data
- Input/output length
- Input tokens
- Instruction tuning
- instructions
- Insufficient data
- Interpretability
- Interpretable model
- IP infringement claims
- Jailbreaking
- Kiro
- Knowledge base
- Label
- Label quality analysis
- Labeled
- Labeled data
- Labeling
- Latency
- Least privilege
- License
- LLM
- LLM-as-a-judge
- Logging
- Loss of customer trust
- Low barrier to entry
- Machine Learning
- Managed API service
- Masking
- Maximum output length
- Meeting business goals
- Memory
- Memory management
- MLOps
- Modality
- Model
- Model complexity
- Model Context Protocol
- Model distillation
- Model monitoring
- Model selection
- Model size
- Model training
- Model type
- Monitoring
- Multi-agent system
- Multiclass
- Multilingual
- Multimodal model
- Negative prompt
- Neural Network
- NLP
- Non-determinism
- Observability
- On-demand
- Open-source models, data, licenses
- Open-source pretrained model
- Operational constraints
- Operational overhead
- Orchestration
- Output filtering and validation
- Output format
- Output tokens
- Output validation
- Overfitting
- Performance
- Performance requirements
- Personalization
- pgvector
- Poisoning
- Policy
- Post-hoc explanation
- Pre-training
- Precision
- Prediction
- Privacy-enhancing technologies
- Production readiness
- Productivity
- Prompt
- Prompt attack protection
- Prompt builder
- Prompt caching
- Prompt-completion pair
- Prompt engineering
- Prompt injection
- Prompt optimization
- Prompt template
- Prompt version
- Proprietary model
- Provisioned throughput
- RAG grounding
- Real-time inference
- Recall
- Recall-Oriented Understudy for Gisting Evaluation
- Recommendation
- Recommendation system
- Redundancy
- Reference text
- Regional coverage
- Regression
- Regulation
- Reinforcement learning
- Repeatable process
- Representativeness
- Responsibility
- Responsiveness
- results
- Retention
- Retraining
- Retrieval-Augmented Generation (RAG)
- Retrieval quality
- Return on Investment
- Reuse
- Review cycle
- Review strategy
- Reward
- Right-sizing
- RLHF
- Robustness
- Rule-based system
- Safety
- SageMaker JumpStart
- SageMaker Model Cards
- Scalability
- Scalable system
- Search
- Security
- Self-hosted API
- Self-supervised
- Semi-
- Sensitive information filters
- Serverless inference
- Shared responsibility model
- Similarity search
- Simplicity
- Single-shot (one-shot)
- Size
- Source citation
- specific
- Specificity and concision
- Speech recognition
- Speed to market
- Stop sequence
- Strands Agents
- Structured
- Subgroup analysis
- Summarization
- Supervised learning
- Supervisor pattern
- Sustainability
- Swarm
- System prompt
- Tabular data
- Task completion
- Task completion rate
- Task engineering
- Team training requirements
- Technical debt
- Temperature
- Text data
- Threat detection
- Time series data
- Token
- Token-based pricing
- Tool
- Tool-call accuracy
- Tool descriptions
- Tool use
- Top K
- Top P
- Toxicity
- Trade-off
- Traditional ML model
- Training
- Transfer learning
- Transformer-based LLM
- Translation
- Transparency
- Transparency of AI decisions
- Transparency standards
- Trustworthiness
- Truthfulness
- Underfitting
- Unlabeled data
- Unstructured
- Unsupervised learning
- User engagement
- User feedback mechanism
- User satisfaction (CSAT)
- Using multiple comments
- Variable
- Variance
- Vector
- Vector database
- Veracity
- Vulnerability management
- Weights
- Word filters
- Workflow
- Workflow orchestration
- Zero-shot
まだありません。項目ページで用語をクリックし、「お気に入り」か「苦手」を選ぶとここに表示されます。