Verfasst von / Written by Sebastian F. Genter
This document provides detailed technical information relevant to our software development processes and product specifications. To ensure clarity and facilitate seamless communication with our international stakeholders, English is the sole language of this documentation. Therefore, a strong command of the English language is considered a fundamental requirement for accessing and applying the technical insights shared within.
Glossary by Topic
- Clustering
- Decision Forests
- ML Fundamentals
- Generative AI
- Image Models
- Language Evaluation
- Metrics
- Recommendation Systems
- Reinforcement Learning
- Responsible AI
- Sequence Models
- TensorFlow
INDEX
A
- A
- ablation
- A/B testing
- accelerator chip
- accuracy
- action
- activation function
- active learning
- AdaGrad
- agent
- agglomerative clustering
- anomaly detection
- AR
- area under the PR curve
- area under the ROC curve
- artificial general intelligence
- artificial intelligence
- attention
- attribute
- attribute sampling
- AUC
- augmented reality
- autoencoder
- automatic evaluation
- automation bias
- AutoML
- autorater evaluation
- auto-regressive model
- auxiliary loss
- average precision at k
- axis-aligned condition
B
- B
- backpropagation
- bagging
- bag of words
- baseline
- batch
- batch inference
- batch normalization
- batch size
- Bayesian neural network
- Bayesian optimization
- Bellman equation
- BERT
- bias (ethics/fairness)
- bias (math)
- bidirectional
- bidirectional language model
- bigram
- binary classification
- binary condition
- binning
- BLEU
- BLEURT
- boosting
- bounding box
- broadcasting
- bucketing
C
- C
- calibration layer
- candidate generation
- candidate sampling
- categorical data
- causal language model
- centroid
- centroid-based clustering
- chain-of-thought prompting
- chat
- checkpoint
- class
- classification model
- classification threshold
- classifier
- class-imbalanced dataset
- clipping
- Cloud TPU
- clustering
- co-adaptation
- collaborative filtering
- concept drift
- condition
- confabulation
- configuration
- confirmation bias
- confusion matrix
- constituency parsing
- contextualized language embedding
- context window
- continuous feature
- convenience sampling
- convergence
- convex function
- convex optimization
- convex set
- convolution
- convolutional filter
- convolutional layer
- convolutional neural network
- convolutional operation
- cost
- co-training
- counterfactual fairness
- coverage bias
- crash blossom
- critic
- cross-entropy
- cross-validation
- CDF
D
- data analysis
- data augmentation
- DataFrame
- data parallelism
- Dataset API (tf.data)
- data set or dataset
- decision boundary
- decision forest
- decision threshold
- decision tree
- decoder
- deep model
- deep neural network
- Deep Q-Network (DQN)
- demographic parity
- denoising
- dense feature
- dense layer
- depth
- depthwise separable convolutional neural network (sepCNN)
- derived label
- device
- differential privacy
- dimension reduction
- dimensions
- direct prompting
- discrete feature
- discriminative model
- discriminator
- disparate impact
- disparate treatment
- distillation
- distribution
- divisive clustering
- downsampling
- DQN
- dropout regularization
- dynamic
- dynamic model
E
- eager execution
- early stopping
- earth mover's distance (EMD)
- edit distance
- Einsum notation
- embedding layer
- embedding space
- embedding vector
- empirical cumulative distribution function (eCDF or EDF)
- empirical risk minimization (ERM)
- encoder
- ensemble
- entropy
- environment
- episode
- epoch
- epsilon greedy policy
- equality of opportunity
- equalized odds
- Estimator
- evals
- evaluation
- example
- experience replay
- experimenter's bias
- exploding gradient problem
F
- F1
- factuality
- fairness constraint
- fairness metric
- false negative (FN)
- false negative rate
- false positive (FP)
- false positive rate (FPR)
- feature
- feature cross
- feature engineering
- feature extraction
- feature importances
- feature set
- feature spec
- feature vector
- featurization
- federated learning
- feedback loop
- feedforward neural network (FFN)
- few-shot learning
- few-shot prompting
- Fiddle
- fine-tuning
- Flax
- Flaxformer
- forget gate
- fraction of successes
- full softmax
- fully connected layer
- function transformation
G
- GAN
- Gemini
- Gemini models
- generalization
- generalization curve
- generalized linear model
- generated text
- generative adversarial network (GAN)
- generative AI
- generative model
- generator
- gini impurity
- golden dataset
- golden response
- GPT
- gradient
- gradient accumulation
- gradient boosted (decision) trees (GBT)
- gradient boosting
- gradient clipping
- gradient descent
- graph
- graph execution
- greedy policy
- groundedness
- ground truth
- group attribution bias
H
I
- i.i.d.
- image recognition
- imbalanced dataset
- implicit bias
- imputation
- incompatibility of fairness metrics
- in-context learning
- independently and identically distributed (i.i.d)
- individual fairness
- inference
- inference path
- information gain
- in-group bias
- input generator
- input layer
- in-set condition
- instance
- instruction tuning
- interpretability
- inter-rater agreement
- intersection over union (IoU)
- IoU
- item matrix
- items
- iteration
J
K
L
- L0 regularization
- L1 loss
- L1 regularization
- L2 loss
- L2 regularization
- label
- labeled example
- label leakage
- lambda
- LaMDA
- landmarks
- language model
- large language model
- latent space
- layer
- Layers API (tf.layers)
- leaf
- LIT
- learning rate
- least squares regression
- Levenshtein Distance
- linear
- linear model
- linear regression
- LIT
- LLM
- LLM evaluations (evals)
- logistic regression
- logits
- Log Loss
- log-odds
- Long Short-Term Memory (LSTM)
- LoRA
- loss
- loss aggregator
- loss curve
- loss function
- loss surface
- Low-Rank Adaptability (LoRA)
- LSTM
M
- machine learning
- machine translation
- majority class
- Markov decision process (MDP)
- Markov property
- masked language model
- matplotlib
- matrix factorization
- Mean Absolute Error (MAE)
- mean average precision at k (mAP@k)
- Mean Squared Error (MSE)
- mesh
- meta-learning
- metric
- Metrics API (tf.metrics)
- mini-batch
- mini-batch stochastic gradient descent
- minimax loss
- minority class
- mixture of experts
- ML
- MMIT
- MNIST
- modality
- model
- model capacity
- model cascading
- model parallelism
- model router
- model training
- MOE
- Momentum
- MT
- multi-class classification
- multi-class logistic regression
- multi-head self-attention
- multimodal instruction-tuned
- multimodal model
- multinomial classification
- multinomial regression
- multitask
N
- NaN trap
- natural language processing
- natural language understanding
- negative class
- negative sampling
- Neural Architecture Search (NAS)
- neural network
- neuron
- N-gram
- NLP
- NLU
- node (decision tree)
- node (neural network)
- node (TensorFlow graph)
- noise
- non-binary condition
- nonlinear
- non-response bias
- nonstationarity
- no one right answer (NORA)
- NORA
- normalization
- novelty detection
- numerical data
- NumPy
O
- objective
- objective function
- oblique condition
- offline
- offline inference
- one-hot encoding
- one-shot learning
- one-shot prompting
- one-vs.-all
- online
- online inference
- operation (op)
- Optax
- optimizer
- out-group homogeneity bias
- outlier detection
- outliers
- out-of-bag evaluation (OOB evaluation)
- output layer
- overfitting
- oversampling
P
- packed data
- pandas
- parameter
- parameter-efficient tuning
- Parameter Server (PS)
- parameter update
- partial derivative
- participation bias
- partitioning strategy
- pass at k (pass@k)
- Pax
- perceptron
- performance
- permutation variable importances
- perplexity
- pipeline
- pipelining
- pjit
- PLM
- pmap
- policy
- pooling
- positional encoding
- positive class
- post-processing
- post-trained model
- PR AUC (area under the PR curve)
- Praxis
- precision
- precision at k (precision@k)
- precision-recall curve
- prediction
- prediction bias
- predictive ML
- predictive parity
- predictive rate parity
- preprocessing
- pre-trained model
- pre-training
- prior belief
- probabilistic regression model
- probability density function
- prompt
- prompt-based learning
- prompt design
- prompt engineering
- prompt tuning
- proxy (sensitive attributes)
- proxy labels
- pure function
Q
R
- RAG
- random forest
- random policy
- rank (ordinality)
- rank (Tensor)
- ranking
- rater
- recall
- recall at k (recall@k)
- recommendation system
- Rectified Linear Unit (ReLU)
- recurrent neural network
- reference text
- regression model
- regularization
- regularization rate
- reinforcement learning (RL)
- Reinforcement Learning from Human Feedback (RLHF)
- ReLU
- replay buffer
- replica
- reporting bias
- representation
- re-ranking
- retrieval-augmented generation (RAG)
- return
- reward
- ridge regularization
- RNN
- ROC (receiver operating characteristic) Curve
- role prompting
- root
- root directory
- Root Mean Squared Error (RMSE)
- rotational invariance
- ROUGE
- ROUGE-L
- ROUGE-N
- ROUGE-S
- R-squared
S
- sampling bias
- sampling with replacement
- SavedModel
- Saver
- scalar
- scaling
- scikit-learn
- scoring
- selection bias
- self-attention
- self-supervised learning
- self-training
- semi-supervised learning
- sensitive attribute
- sentiment analysis
- sequence model
- sequence-to-sequence task
- serving
- shape (Tensor)
- shard
- shrinkage
- sigmoid function
- similarity measure
- single program / multiple data (SPMD)
- size invariance
- sketching
- skip-gram
- softmax
- soft prompt tuning
- sparse feature
- sparse representation
- sparse vector
- sparsity
- spatial pooling
- split
- splitter
- SPMD
- squared hinge loss
- squared loss
- staged training
- state
- state-action value function
- static
- static inference
- stationarity
- step
- step size
- stochastic gradient descent (SGD)
- stride
- structural risk minimization (SRM)
- subsampling
- subword token
- summary
- supervised machine learning
- synthetic feature
T
- T5
- T5X
- tabular Q-learning
- target
- target network
- task
- temperature
- temporal data
- Tensor
- TensorBoard
- TensorFlow
- TensorFlow Playground
- TensorFlow Serving
- Tensor Processing Unit (TPU)
- Tensor rank
- Tensor shape
- Tensor size
- TensorStore
- termination condition
- test
- test loss
- test set
- text span
- tf.Example
- tf.keras
- threshold (for decision trees)
- time series analysis
- timestep
- token
- top-k accuracy
- tower
- toxicity
- TPU
- TPU chip
- TPU device
- TPU node
- TPU Pod
- TPU resource
- TPU slice
- TPU type
- TPU worker
- training
- training loss
- training-serving skew
- training set
- trajectory
- transfer learning
- Transformer
- translational invariance
- trigram
- true negative (TN)
- true positive (TP)
- true positive rate (TPR)