CONCEPT ENCYCLOPEDIA
Follow the connections.
1,345 permanent reference pages · 21 expanded guides. Every reference page preserves its original definition, example, caveat, context and sources. The remaining detailed expansions are visibly pending.
Start with an expanded guide
- Machine learning
- Deep learning
- Large language model
- Small language model
- Transformer
- Self-attention
- Retrieval-augmented generation
- Embedding
- Token
- Overfitting
- Gradient descent
- Fine-tuning
- Specification gaming
- Reward hacking
- Goodhart's law
- Goal misgeneralization
- Deceptive alignment
- Shutdown avoidance
- Activation-based monitoring
- Mechanistic interpretability
- Red teaming
All concepts, A–Z
- A/B test
- Ablation study
- Abliteration
- Abstention
- Acceptance rate
- Access control
- Accuracy
- Acquisition function
- Action
- Action space
- Activation
- Activation checkpointing
- Activation function
- Activation patching
- Activation quantization
- Activation steering
- Activation-based monitoring
- Active learning
- Active parameters
- Actor
- Actor-critic
- AdaBoost
- Adafactor
- AdaGrad
- Adam
- AdamW
- Adapter merging
- Adapter tuning
- Adaptive optimizer
- Adjusted mutual information
- Adjusted Rand index
- Advantage
- Adversarial evaluation
- Adversarial example
- Adversarial loss
- Adversarial robustness
- Adversarial training
- Affinity propagation
- Agent
- Agent containment
- Agent evaluation
- Agent memory
- Agentic workflow
- Agglomerative clustering
- AI alignment
- AI safety
- Aleatoric uncertainty
- Algorithm
- ALiBi
- Alignment faking
- Alignment tax
- All-gather
- All-reduce
- Alternative hypothesis
- Annotation
- Anomaly detection
- Answer normalization
- Answer relevance
- Approximate nearest-neighbor search
- Architecture
- ARIMA
- Arithmetic intensity
- Artificial intelligence
- Assistant-only loss
- Asymmetric quantization
- Attack surface
- Attention
- Attention distillation
- Attention head
- Attention map
- Attention mask
- Attention score
- Attention weight
- Attribution graph
- Autocorrelation
- Autoencoder
- Automatic differentiation
- Automatic speech recognition
- Autoregression
- Auxiliary loss
- Average pooling
- Average precision
- AWQ
- Backbone
- Backdoor
- Backpropagation
- Backpropagation through time
- Bag of words
- Bagging
- Balanced accuracy
- Ball tree
- Bandwidth
- Base model
- Base optimizer
- Baseline
- Batch inference
- Batch learning
- Batch normalization
- Batch size
- Bayes' theorem
- Bayesian inference
- Bayesian linear regression
- Bayesian neural network
- Bayesian optimization
- Beam search
- Beam width
- Behavioral cloning
- Belief state
- Bellman equation
- Benchmark
- Bernoulli distribution
- Bernoulli naive Bayes
- BERT
- BERTScore
- Best-of-N sampling
- Bfloat16
- Bi-encoder
- Bias audit
- Bias parameter
- Bias-variance tradeoff
- Bidirectional context
- Bidirectional RNN
- Binary classification
- Binary cross-entropy
- Binomial distribution
- Bits per character
- Blending
- BLEU
- BM25
- Bonferroni correction
- Boosting
- Bootstrap
- Bootstrapping (RL)
- BOS token
- Bottleneck
- Bounding box
- Brier score
- Broadcasting
- Byte fallback
- Byte-level tokenization
- Byte-pair encoding
- Calibration
- Calibration dataset
- Canary deployment
- Capability evaluation
- Capability failure
- Catastrophic forgetting
- CatBoost
- Categorical cross-entropy
- Categorical distribution
- Categorical feature
- Causal attention
- Causal inference
- Causal intervention
- Causal language modeling
- Causal mask
- Causal mediation analysis
- Causation
- Centroid
- Chain rule
- Chain-of-thought prompting
- Change of variables
- Channel
- Character error rate
- Chat template
- Checkpoint
- Chosen response
- Chunk overlap
- Chunked prefill
- Chunking
- Circuit
- Circuit tracing
- Citation attribution
- Class imbalance
- Class weight
- Classification
- Classifier-free guidance
- CLIP
- Clustering
- Codebook
- Cohen's kappa
- Cold-start problem
- Collaborative filtering
- Collective communication
- Completion-only training
- Computation graph
- Computational complexity
- Compute kernel
- Compute-optimal training
- Computer vision
- Concept drift
- Condition number
- Conditional independence
- Conditional probability
- Conditioning
- Confidence interval
- Confirmation bias
- Conformal prediction
- Confounding
- Confusion matrix
- Connectionist temporal classification
- Constitutional AI
- Constrained decoding
- Content-based filtering
- Context compression
- Context engineering
- Context extension
- Context parallelism
- Context relevance
- Context window
- Contextual bandit
- Contextual embedding
- Continual learning
- Continued instruction tuning
- Continued pretraining
- Continuous bag of words
- Continuous batching
- Continuous feature
- Continuous normalizing flow
- Contrastive learning
- Contrastive loss
- Control group
- Convergence
- Convex function
- Convex optimization
- Convolution
- Convolution kernel
- Convolutional neural network
- Corpus
- Correlation
- Corrigibility
- Cosine decay
- Cosine similarity
- Counterfactual
- Counterfactual explanation
- Covariance
- Covariate shift
- CPU offloading
- Credible interval
- Credit assignment
- Critic
- Cross-attention
- Cross-encoder
- Cross-entropy
- Cross-fitting
- Cross-modal alignment
- Cross-validation
- Cryptographic hash
- CSR format
- CUDA
- Cumulative distribution function
- Curriculum learning
- Curse of dimensionality
- CutMix
- Data augmentation
- Data cleaning
- Data curation
- Data curriculum
- Data deduplication
- Data drift
- Data extraction
- Data filtering
- Data leakage
- Data lineage
- Data minimization
- Data mixture
- Data parallelism
- Data poisoning
- Data preprocessing
- Data provenance
- Data quality
- Data split
- Data versioning
- Dataset
- Dataset card
- DBSCAN
- DCG
- Dead neuron
- Deception
- Deceptive alignment
- Decision boundary
- Decision forest
- Decision threshold
- Decision tree
- Decode phase
- Decoder
- Decoder-only model
- Decoding
- Deep learning
- Deep Q-network
- Defense in depth
- Demographic parity
- Dendrogram
- Denoising
- Denoising autoencoder
- Denoising diffusion implicit model
- Denoising diffusion probabilistic model
- Denoising objective
- Dense model
- Dense retrieval
- Density estimation
- Density-based clustering
- Depthwise convolution
- Depthwise separable convolution
- Determinism
- Detokenization
- Dice coefficient
- Dice loss
- Dictionary learning
- Differential privacy
- Diffusion model
- Diffusion sampler
- Dilated convolution
- Dimensionality reduction
- Disaggregated serving
- Discount factor
- Discretization
- Discriminative model
- Discriminator
- Distance metric
- Distillation loss
- Distillation temperature
- Distributed data parallel
- Distributed learning
- Distributed training
- Distribution shift
- Diversity
- Document index
- Document understanding
- Domain adaptation
- Domain generalization
- Domain-adaptive pretraining
- DoRA
- Dot product
- Double descent
- Double quantization
- DP-SGD
- DPO
- Draft model
- DropConnect
- Dropout
- Dynamic batching
- Dynamic padding
- Dynamic programming
- Early stopping
- Edge AI
- Edit distance
- Effect size
- Effective batch size
- Effective horizon
- Effective sample size
- Eigenvalue
- Eigenvector
- Elastic net
- Elastic weight consolidation
- ELU
- Embedding
- Embedding dimension
- Embedding matrix
- Embedding space
- Empirical risk
- Empirical risk minimization
- Encoder
- Encoder-decoder model
- Encoder-only model
- End-to-end learning
- Ensemble learning
- Entity linking
- Entropy
- Entropy bonus
- Environment
- EOS token
- Episode
- Episodic memory
- Epistemic uncertainty
- Epoch
- Epsilon-greedy
- Equal opportunity (fairness)
- Equalized odds
- Euclidean distance
- Evaluation awareness
- Evaluation variance
- Evaluator bias
- Evaluator integrity
- Evidence lower bound
- Exact match
- Expectation
- Expectation-maximization
- Expected calibration error
- Expected risk
- Experiment tracking
- Expert
- Expert capacity
- Expert parallelism
- Explainability
- Explained variance
- Exploration-exploitation tradeoff
- Exponential decay
- Exposure bias
- Expressivity
- External validity
- Extra Trees
- Extrapolation
- F-beta score
- F1 score
- Factor analysis
- Fairness
- Faithfulness
- Fake quantization
- False discovery rate
- False negative
- False negative rate
- False positive
- False positive rate
- Feature
- Feature attribution
- Feature distillation
- Feature engineering
- Feature extraction
- Feature map
- Feature scaling
- Feature selection
- Feature splitting
- Feature store
- Feature vector
- Federated learning
- Feed-forward network
- Few-shot learning
- Few-shot prompting
- Fill-in-the-middle
- Fine-tuning
- Fisher information
- FlashAttention
- Float16
- Float32
- Float8
- Floating-point operation count
- Floating-point operations per second
- Flow matching
- Focal loss
- Forecasting
- Forward-mode autodiff
- Foundation model
- Frequency penalty
- Frobenius norm
- Full attention
- Full fine-tuning
- Fully sharded data parallel
- Function calling
- Gated linear unit
- Gating mechanism
- Gaussian mixture model
- Gaussian naive Bayes
- Gaussian process
- Gaussian process regression
- GELU
- Generalization
- Generalized additive model
- Generalized advantage estimation
- Generalized linear model
- Generative adversarial network
- Generative AI
- Generative model
- Generator
- Gini impurity
- Global attention
- Global average pooling
- Global minimum
- Goal misgeneralization
- Goodhart's law
- GPTQ
- GPU
- Grad-CAM
- Gradient
- Gradient accumulation
- Gradient boosting
- Gradient clipping
- Gradient descent
- Gradient estimator
- Gradient explosion
- Gradient noise
- Gradient norm
- Gradient penalty
- Gradient-boosted decision trees
- Grammar-constrained decoding
- Graph attention network
- Graph convolutional network
- Graph Laplacian
- Graph neural network
- GraphRAG
- Greedy decoding
- Grid search
- Grokking
- Grounding
- Group cross-validation
- Group normalization
- Group split
- Grouped-query attention
- Groupwise quantization
- GRPO
- GRU
- Hadamard product
- Hallucination
- Hard targets
- Hard-negative mining
- Harmlessness
- Hashing trick
- HDBSCAN
- He initialization
- Head dimension
- Helpfulness
- Hessian
- Heuristic
- Hidden layer
- Hidden size
- Hidden state
- Hierarchical clustering
- Hierarchical retrieval
- Hinge loss
- Histogram-based boosting
- Hit rate
- HNSW
- Honesty
- Huber loss
- Hugging Face Transformers
- Human evaluation
- Human oversight
- Human-in-the-loop
- HumanEval
- Hybrid architecture
- Hybrid retrieval
- HyDE
- Hyperband
- Hyperparameter
- Hyperparameter optimization
- Hypothesis class
- IID
- Image classification
- Image-to-image generation
- Imitation learning
- Implicit regularization
- Importance sampling
- Importance weighting
- Imputation
- In-context learning
- Incremental learning
- Independence
- Independent component analysis
- Indirect prompt injection
- Individual conditional expectation
- Individual fairness
- Induction head
- Inductive bias
- Inference
- Inference endpoint
- Infilling
- InfoNCE
- Information gain
- Information retrieval
- Inner alignment
- Inpainting
- Input layer
- Instance normalization
- Instance segmentation
- Instruction hierarchy
- Instruction tuning
- Instruction-tuned model
- Instrumental convergence
- Integrated gradients
- Inter-annotator agreement
- Inter-token latency
- Interaction feature
- Internal validity
- Interpolation
- Interpolation threshold
- Interpretability
- Intersection over union
- Inverse matrix
- Inverse reinforcement learning
- Inverted file index
- Inverted index
- Invertible neural network
- Isolation forest
- Isomap
- Isotonic regression
- Jacobian
- Jacobian-vector product
- Jailbreak
- Jensen-Shannon divergence
- Joint probability
- K-fold cross-validation
- K-means
- K-means++
- K-medoids
- K-nearest neighbors
- KD-tree
- Kernel
- Kernel approximation
- Kernel density estimation
- Kernel fusion
- Kernel PCA
- Kernel ridge regression
- Kernel trick
- Key
- Key-value cache
- KL penalty
- Knowledge distillation
- Knowledge graph
- Knowledge retention
- Kullback-Leibler divergence
- KV-cache quantization
- L-BFGS
- L1 loss
- L1 norm
- L1 regularization
- L2 loss
- L2 norm
- L2 regularization
- Label
- Label noise
- Label shift
- Label smoothing
- Lag feature
- Language model
- Large language model
- Lasso
- Late interaction
- Latency
- Latent diffusion model
- Latent Dirichlet allocation
- Latent space
- Latent variable
- Layer
- Layer normalization
- Leaf node
- Leaky ReLU
- Learning curve
- Learning rate
- Learning to rank
- Learning-rate schedule
- Least privilege
- Leave-one-out cross-validation
- Length bias
- Length bucketing
- Length penalty
- LightGBM
- Likelihood
- LIME
- Linear attention
- Linear decay
- Linear discriminant analysis
- Linear layer
- Linear probe
- Linear regression
- Linear system
- Link prediction
- Linkage criterion
- Lipschitz continuity
- Listwise ranking loss
- LLM-as-a-judge
- LLMOps
- Load-balancing loss
- Local minimum
- Local outlier factor
- Locality-sensitive hashing
- Locally linear embedding
- Log loss
- Log transformation
- Log-normal distribution
- Log-odds
- Log-softmax
- Log-sum-exp
- Logistic regression
- Logit
- Logit distillation
- Logit lens
- Logits processor
- Long-context evaluation
- Longest common subsequence
- LoRA alpha
- LoRA rank
- Loss function
- Loss landscape
- Loss masking
- Loss scaling
- Lost-in-the-middle effect
- Low-bit inference
- Low-rank adaptation
- Low-rank approximation
- LSTM
- Machine learning
- Macro averaging
- Majority voting
- Mamba
- Manifold learning
- Margin
- Marginal probability
- Markov chain
- Markov chain Monte Carlo
- Markov decision process
- Markov property
- Masked language modeling
- Matrix
- Matrix factorization
- Matrix multiplication
- Matrix rank
- Matthews correlation coefficient
- Max pooling
- Maximum a posteriori estimation
- Maximum likelihood estimation
- Maximum marginal relevance
- Mean absolute error
- Mean absolute percentage error
- Mean average precision
- Mean reciprocal rank
- Mean shift
- Mean squared error
- Mechanistic interpretability
- Median absolute error
- Medoid
- Mel spectrogram
- Membership inference
- Memorization
- Memory bandwidth
- Mesa-objective
- Mesa-optimization
- Mesa-optimizer
- Message passing
- Meta-learning
- Metadata filtering
- Metric learning
- Micro averaging
- Microbatch
- Min-max scaling
- Min-p sampling
- MinHash
- Mini-batch
- Minimax optimization
- Missing value
- Missingness indicator
- Mixed-precision training
- Mixture of experts
- Mixture of models
- Mixup
- ML pipeline
- MLflow
- MLOps
- MMLU
- Mode collapse
- Model
- Model artifact
- Model capacity
- Model card
- Model compression
- Model Context Protocol
- Model deployment
- Model extraction
- Model inversion
- Model merging
- Model monitoring
- Model parallelism
- Model registry
- Model selection
- Model serving
- Model versioning
- Model-based reinforcement learning
- Model-free reinforcement learning
- Momentum
- Monosemanticity
- Monte Carlo estimation
- Multi-agent reinforcement learning
- Multi-agent system
- Multi-armed bandit
- Multi-head attention
- Multi-head latent attention
- Multi-objective optimization
- Multi-query attention
- Multi-query retrieval
- Multi-token prediction
- Multiclass classification
- Multidimensional scaling
- Multihead network
- Multilabel classification
- Multilayer perceptron
- Multimodal model
- Multinomial distribution
- Multinomial naive Bayes
- Multiple imputation
- Multiple testing
- Multitask learning
- Mutual information
- N-gram
- Naive Bayes
- Named entity recognition
- Natural gradient
- NDCG
- Near-duplicate detection
- Nearest centroid classifier
- Nearest-neighbor search
- Negative log-likelihood
- Negative sampling
- Nested cross-validation
- Nesterov momentum
- Neural network
- Neural ordinary differential equation
- Neuron
- Newton's method
- NF4
- No-free-lunch theorem
- Node embedding
- Noise schedule
- Noise-contrastive estimation
- Non-maximum suppression
- Nonconvex optimization
- Nonnegative matrix factorization
- Nonstationarity
- Norm
- Normal distribution
- Normalization
- Normalizing flow
- Novelty detection
- Null hypothesis
- Numerical overflow
- Numerical stability
- Numerical underflow
- Nyström approximation
- Object detection
- Objective function
- Observability
- Observation
- Odds ratio
- Off-policy distillation
- Off-policy evaluation
- Off-policy learning
- Offline distillation
- Offline reinforcement learning
- On-device inference
- On-policy distillation
- On-policy learning
- One-class SVM
- One-hot encoding
- One-shot learning
- Online distillation
- Online inference
- Online learning
- OOD detection
- Optical character recognition
- Optimization
- Optimizer
- Optimizer state
- Orchestrator
- Ordinal encoding
- Out-of-bag evaluation
- Out-of-distribution
- Out-of-vocabulary
- Outcome reward model
- Outcome supervision
- Outer alignment
- Outer product
- Outlier
- Output layer
- Over-smoothing
- Overfitting
- Overrefusal
- Oversampling
- P-value
- PAC learning
- Padding mask
- PagedAttention
- Paired evaluation
- Pairwise ranking loss
- Panoptic segmentation
- Parameter
- Parameter count
- Parameter-efficient fine-tuning
- Parent-document retrieval
- Pareto frontier
- Partial dependence plot
- Partial derivative
- Partially observable Markov decision process
- Pass@k
- Patch embedding
- PEFT library
- Per-channel quantization
- Per-tensor quantization
- Perceptual hash
- Perceptual loss
- Permutation importance
- Permutation test
- Perplexity
- Pinball loss
- Pipeline parallelism
- Planning
- Platt scaling
- Pointwise convolution
- Poisson distribution
- Poisson regression
- Policy
- Policy collapse
- Policy gradient
- Policy iteration
- Polynomial features
- Polynomial kernel
- Polynomial regression
- Polysemanticity
- Pooling
- Position bias
- Position embedding
- Positional encoding
- Positional feature
- Positive semidefinite matrix
- Post-norm Transformer
- Post-training
- Post-training quantization
- Posterior
- Posterior collapse
- Potential-based reward shaping
- Power law
- Power-seeking
- PPO
- PR AUC
- Pre-activation
- Pre-norm Transformer
- Pre-tokenization
- Precision
- Precision at k
- Precision-recall curve
- Preconditioning
- Prediction
- Prediction head
- Prediction interval
- Preference dataset
- Preference optimization
- Prefill
- Prefix caching
- Prefix tuning
- PReLU
- Presence penalty
- Pretraining
- Principal component analysis
- Prior
- Prioritized experience replay
- Privacy accounting
- Privacy budget
- Privacy leakage
- Probability density
- Probability distribution
- Probability mass function
- Probe
- Process reward model
- Process supervision
- Product quantization
- Prompt
- Prompt chaining
- Prompt engineering
- Prompt ensembling
- Prompt injection
- Prompt sensitivity
- Prompt tuning
- Prompt versioning
- Proxy objective
- Pruning
- Pseudo-labeling
- Pseudoinverse
- PyTorch
- Q function
- Q-learning
- QLoRA
- Quadratic complexity
- Quadratic discriminant analysis
- Quantile
- Quantile regression
- Quantization
- Quantization scale
- Quantization-aware training
- Quasi-Newton method
- Query
- Query expansion
- Query rewriting
- R-squared
- Radial basis function kernel
- Random forest
- Random Fourier features
- Random search
- Random seed
- Random variable
- Randomized experiment
- Rank-stabilized LoRA
- Ranking
- RANSAC
- ReAct
- Reasoning trace
- Recall
- Recall at k
- Receptive field
- Reciprocal rank fusion
- Recommendation
- Reconstruction loss
- Rectified flow
- Recurrent neural network
- Recursive feature elimination
- Red teaming
- Reduce-scatter
- Reference model
- Reflection
- Refusal
- Refusal direction
- Regression
- Regularization
- Regularization loss
- REINFORCE
- Reinforcement learning
- Rejected response
- Rejection-sampling fine-tuning
- Relative position encoding
- Reliability diagram
- ReLU
- Reparameterization trick
- Repetition penalty
- Replanning
- Replay buffer
- Replication
- Representation learning
- Reproducibility
- Reranking
- Resampling
- Residual connection
- Residual network
- Residual stream
- Retrieval corpus
- Retrieval precision
- Retrieval recall
- Retrieval-augmented generation
- Retriever
- Return
- Reverse-mode autodiff
- Reward
- Reward function
- Reward hacking
- Reward model
- Reward model overoptimization
- Reward shaping
- Reward tampering
- Ridge regression
- RLAIF
- RLHF
- RLVR
- RMS normalization
- RMSProp
- Robust loss
- Robust optimization
- Robust regression
- Robust scaling
- Robustness
- ROC AUC
- ROC curve
- Rollback
- Root mean squared error
- Root node
- Rotary position embedding
- ROUGE
- Router
- Saddle point
- Safety case
- Safety evaluation
- Safety training
- Saliency map
- Sample
- Sample complexity
- Sample size
- Sample weight
- Sampling
- Sampling bias
- Sampling variance
- Sandbagging
- Sandbox
- SARSA
- Scalable oversight
- Scalar
- Scaled dot-product attention
- Scaling law
- Scheduled sampling
- Scheming
- scikit-learn
- Score function
- Score matching
- Score-based generative model
- Search budget
- Seasonality
- Security evaluation
- Selection bias
- Selection pressure
- Selective prediction
- Selective state-space model
- Self-attention
- Self-consistency
- Self-correction
- Self-distillation
- Self-paced learning
- Self-supervised learning
- Self-training
- Semantic chunking
- Semantic memory
- Semantic search
- Semantic segmentation
- Semi-supervised learning
- Sentence embedding
- SentencePiece
- Sequence length
- Sequence packing
- Sequence padding
- Sequence parallelism
- Sequence-level distillation
- Sequence-to-sequence
- Shadow deployment
- SHAP
- Shapley value
- Shortcut learning
- Shutdown avoidance
- Shutdownability
- Sigmoid
- Signal-to-noise ratio
- Silhouette score
- SiLU
- Singular value decomposition
- Sinusoidal positional encoding
- Situational awareness
- Skip-gram
- Sleeper agent (model)
- Sliding-window attention
- Small language model
- SMOTE
- Soft prompt
- Soft targets
- Soft-margin SVM
- Softmax
- Span corruption
- Sparse attention
- Sparse autoencoder
- Sparse coding
- Sparse feature
- Sparse kernel
- Sparse matrix
- Sparse model
- Sparse retrieval
- Sparse reward
- Spatial padding
- Special token
- Specification gaming
- Specificity
- Spectral clustering
- Spectral norm
- Spectral normalization
- Spectrogram
- Speculative decoding
- Split criterion
- Spurious correlation
- Squared hinge loss
- Stacking
- Standard deviation
- Standard error
- Standardization
- State
- State-space model
- Stationarity
- Statistical bias
- Statistical power
- Statistical significance
- Steering vector
- Stochastic depth
- Stochastic gradient descent
- Stop sequence
- Stop-gradient
- Straight-through estimator
- Stratified K-fold
- Stratified split
- Stride
- Structural risk minimization
- Structured output
- Structured pruning
- Structured sparsity
- Student model
- Subagent
- Subgroup evaluation
- Subword regularization
- Subword token
- Successive halving
- Superposition
- Supervised fine-tuning
- Supervised learning
- Support vector
- Support vector machine
- Support vector regression
- Surrogate model
- Survivorship bias
- SwiGLU
- Sycophancy
- Symmetric MAPE
- Symmetric quantization
- Synthetic data
- System prompt
- t-SNE
- T5
- Tanh
- Target encoding
- Target leakage
- Target model
- Target network
- Task arithmetic
- Task vector
- Task-adaptive pretraining
- TD error
- Teacher forcing
- Teacher model
- Temperature
- Temperature scaling
- Temporal cross-validation
- Temporal split
- Temporal-difference learning
- Tensor
- Tensor parallelism
- Tensor rank
- Tensor shape
- Terminal state
- Test set
- Test-time compute
- Test-time scaling
- Text normalization
- Text-to-speech
- TF-IDF
- Threat model
- Throughput
- Time series
- Time to first token
- TinyML
- Token
- Token budget
- Token count
- Token dropping
- Token ID
- Token-to-parameter ratio
- Tokenization
- Tokenizer
- Tool result
- Tool schema
- Tool use
- Top-k accuracy
- Top-k routing
- Top-k sampling
- Top-p sampling
- Topic modeling
- Total parameters
- Trace
- Train-test contamination
- Training
- Training instability
- Training pipeline
- Training schedule
- Training set
- Training step
- Training-serving skew
- Trajectory
- Transfer learning
- Transformer
- Transformer block
- Transpose
- Transposed convolution
- Tree depth
- Tree of thoughts
- Trend
- Triplet loss
- TRL
- True negative
- True positive
- Truncated backpropagation through time
- Truncation
- Truthfulness
- Type I error
- Type II error
- Typical sampling
- U-Net
- UMAP
- Uncertainty estimation
- Uncertainty sampling
- Underfitting
- Undersampling
- Unigram tokenizer
- Unknown token
- Unstructured pruning
- Unsupervised learning
- Upsampling
- User prompt
- Validation set
- Value
- Value function
- Value iteration
- Vanishing gradient
- Variance
- Variational autoencoder
- Variational inference
- VC dimension
- Vector
- Vector database
- Vector field
- Vector index
- Vector search
- Vector-Jacobian product
- Verifier
- Vision transformer
- Vision-language model
- Visual question answering
- vLLM
- Vocabulary
- Vocabulary size
- Vocoder
- Voting ensemble
- VRAM
- Warm restarts
- Warmup
- Wasserstein distance
- Wasserstein GAN
- Weak supervision
- Weak-to-strong generalization
- Weight
- Weight decay
- Weight initialization
- Weight tying
- Weight-only quantization
- Weighted averaging
- Winsorization
- Wireheading
- Word embedding
- Word error rate
- Word2Vec
- WordPiece
- Workflow
- Working memory
- World model
- Worst-group accuracy
- Xavier initialization
- XGBoost
- ZeRO
- Zero point
- Zero-shot learning
- Zero-shot prompting