Training Glossary
Browse 117 training terms defined in plain English, from the cultural dictionary of computing.
117 Training Terms
- Activation Function
- A mathematical function applied to a neural unit's output to introduce nonlinearity into the model. It influences how models are trained, evaluated, or served,...
- Active Learning
- A machine learning training concept related to active learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Adaptive Learning Rate
- A machine learning training concept related to adaptive learning rate and how model parameters are learned or stabilized. It influences how models are trained,...
- Adversarial Training
- A machine learning training concept related to adversarial training and how model parameters are learned or stabilized. It influences how models are trained,...
- AI Training
- The process of adjusting a model's parameters using data so it learns patterns or capabilities relevant to a task or domain. In product discussions, the term...
- Batch Size
- A machine learning training concept related to batch size and how model parameters are learned or stabilized. It influences how models are trained, evaluated,...
- Chinchilla Scaling
- A scaling-law result emphasizing that, for a given compute budget, model size and training data should be balanced rather than simply making models larger with...
- Compute Optimal Training
- A training regime chosen to get the best model quality for a fixed compute budget by balancing model size and data volume appropriately.
- Constitutional AI Method
- An alignment method in which an AI system is guided by an explicit set of principles or rules, often using self-critique and revision to improve responses...
- Context Distillation
- A technique for training or adapting models so they internalize behaviors that would otherwise require large context at inference time. The goal is to compress...
- CTF
- CTF is a capture-the-flag exercise where participants solve security challenges to practice exploitation, analysis, and defense. Security teams use it to...
- Curriculum Learning
- A machine learning training concept related to curriculum learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Cyber Range
- Cyber Range is a controlled environment for realistic security training, attack simulation, and defensive exercises. Security teams use it to enforce trust,...
- Deep Learning
- A machine learning training concept related to deep learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Deep Reinforcement Learning
- A machine learning training concept related to deep reinforcement learning and how model parameters are learned or stabilized. It influences how models are...
- distillation
- Distillation (or knowledge distillation) is a model compression technique in machine learning where a smaller, more efficient student model is trained to...
- Distributed Training
- A machine learning training concept related to distributed training and how model parameters are learned or stabilized. It influences how models are trained,...
- Dropout Detail
- A machine learning training concept related to dropout detail and how model parameters are learned or stabilized. It influences how models are trained,...
- End-to-End Learning
- A machine learning training concept related to end-to-end learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Epoch
- One complete pass through the entire training dataset. Models are typically trained for multiple epochs — enough to learn patterns but not so many that they...
- Expert Iteration
- A training approach in which a stronger search or expert process generates improved targets or guidance that a model then learns to imitate, repeating this...
- Exponential Moving Average
- A smoothed running average that weights recent observations more heavily than older ones. It influences how models are trained, evaluated, or served, and it...
- Fine-Tuning
- The process of taking a pre-trained model and further training it on a smaller, domain-specific dataset to improve performance on a particular task. Like...
- Frozen Embedding
- An embedding layer or embedding representation whose parameters are kept fixed during some later stage of training or adaptation. Freezing embeddings can...
- Full Fine-Tuning
- A machine learning training concept related to full fine-tuning and how model parameters are learned or stabilized. It influences how models are trained,...
- Goal Conditioned
- Describing a model or policy that takes an explicit goal as part of its input so its behavior can adapt to the desired outcome. Goal-conditioned approaches are...
- Gradient
- A machine learning training concept related to gradient and how model parameters are learned or stabilized. It influences how models are trained, evaluated, or...
- Gradient Accumulation
- A machine learning training concept related to gradient accumulation and how model parameters are learned or stabilized. It influences how models are trained,...
- Gradient Checkpointing
- A machine learning training concept related to gradient checkpointing and how model parameters are learned or stabilized. It influences how models are trained,...
- Gradient Clipping
- A machine learning training concept related to gradient clipping and how model parameters are learned or stabilized. It influences how models are trained,...
- Gradient Descent
- A machine learning training concept related to gradient descent and how model parameters are learned or stabilized. It influences how models are trained,...
- Gradient Explosion
- A machine learning training concept related to gradient explosion and how model parameters are learned or stabilized. It influences how models are trained,...
- Gradient-Free Optimization
- Optimization methods that do not rely on computing gradients and instead search using alternatives such as evolutionary strategies, sampling, or black-box...
- Gradient Vanishing
- A machine learning training concept related to gradient vanishing and how model parameters are learned or stabilized. It influences how models are trained,...
- Group Normalization
- A normalization method that standardizes activations within channel groups instead of across the batch. It influences how models are trained, evaluated, or...
- Hyperparameter Sweep
- A systematic run of many training jobs with different hyperparameter settings, such as learning rate or batch size, to find stronger model performance.
- Hyperparameter Tuning Detail
- A machine learning training concept related to hyperparameter tuning detail and how model parameters are learned or stabilized. It influences how models are...
- Imitation Learning
- A machine learning training concept related to imitation learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Incremental Learning
- A machine learning training concept related to incremental learning and how model parameters are learned or stabilized. It influences how models are trained,...
- Instruction Tuning
- A machine learning training concept related to instruction tuning and how model parameters are learned or stabilized. It influences how models are trained,...
- Knowledge Cutoff
- The latest point in time up to which a model's training data is known or assumed to reflect information. Knowledge cutoff matters because models without...
- Layer Normalization
- A normalization method that standardizes activations across features within each example. It influences how models are trained, evaluated, or served, and it...
- Leaky ReLU
- An activation function that keeps a small nonzero slope for negative inputs. It influences how models are trained, evaluated, or served, and it can materially...
- Learning Curriculum
- A machine learning training concept related to learning curriculum and how model parameters are learned or stabilized. It influences how models are trained,...
- Learning Rate
- A machine learning training concept related to learning rate and how model parameters are learned or stabilized. It influences how models are trained,...
- Learning Rate Schedule
- A machine learning training concept related to learning rate schedule and how model parameters are learned or stabilized. It influences how models are trained,...
- Learning Rate Warmup
- A machine learning training concept related to learning rate warmup and how model parameters are learned or stabilized. It influences how models are trained,...
- Loss Function
- Loss Function is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for...
- Loss Landscape
- Loss Landscape is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for...
- Machine Unlearning
- Techniques aimed at removing the influence of specific data from a trained model without fully retraining it from scratch. Machine unlearning is discussed in...
- Model Collapse
- A phenomenon where AI models trained on AI-generated data progressively degrade, losing diversity and accuracy over generations. The ouroboros of AI — models...
- Model Distillation Loss
- The training loss used when a student model learns to match outputs or distributions from a teacher model during distillation.
- Model Fine-Tuning Cost
- The total expense of adapting a base model to a new task, including compute, data preparation, experimentation, and engineering time.
- Neural Network Depth
- The number of layers in a neural network, which affects representational capacity, compute cost, and optimization behavior.
- Neural Scaling Law
- Neural Scaling Law is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and...
- Neural Scaling Laws
- Empirical relationships showing how model performance changes as compute, data, or parameter count increases.
- Next Token Prediction
- The task of predicting the next token in a sequence given the preceding context, which is a core training objective for many language models. Despite its...
- Noise Injection
- The deliberate addition of noise to inputs, activations, gradients, or parameters during training or testing to improve robustness, exploration, or...
- NVIDIA H100
- A high-end NVIDIA data-center GPU widely used for training and serving large AI models.
- Objective Function
- The function a model or training process is trying to optimize, such as minimizing error or maximizing reward. The objective function strongly shapes what the...
- Optimizer
- Optimizer is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for neural...
- Overfitting
- When a model learns the training data too well — including its noise and random fluctuations — and fails to generalize to new, unseen data. The model has...
- Pentest Lab
- A controlled environment built for practicing, validating, or teaching penetration testing without risking production systems. Pentest labs often include...
- Performance Benchmark
- Performance Benchmark is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before...
- Perplexity Detail
- Perplexity Detail is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before release...
- Phishing Simulation
- A controlled internal exercise that sends realistic but benign phishing-style messages to measure and improve user awareness, reporting, and response habits....
- Pipeline Parallelism
- Pipeline Parallelism is a systems strategy for splitting model computation or parameters across devices. It is commonly used for training and serving workloads...
- Precision Detail
- Precision Detail is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before release or...
- Preference Data
- Data capturing which outputs humans prefer among alternatives, often used to train ranking models, reward models, or alignment systems. Preference data is...
- Process Reward Model
- A reward model that evaluates the quality of intermediate reasoning steps or process traces rather than scoring only the final answer. Process reward models...
- Ranking Loss
- Ranking Loss is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for neural...
- Recall Detail
- Recall Detail is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before release or...
- Refusal Training
- Training or tuning methods intended to improve when and how an AI system refuses unsafe, disallowed, or unsupported requests. Refusal training is part of...
- Regularization
- Regularization is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for...
- Replay Buffer
- A stored collection of past experiences or transitions used in reinforcement learning so training can reuse prior data instead of relying only on the newest...
- Reward Signal
- The signal used to tell a learning system how good or bad an action, output, or trajectory was. Reward signals drive optimization, so poorly specified signals...
- RLHF
- Reinforcement Learning from Human Feedback — a training technique where human evaluators rank model outputs by quality, and the model is trained to produce...
- ROC Curve
- ROC Curve is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before release or model...
- Safety Training
- Training or tuning intended to improve safe behavior, refusal quality, and compliance with policies or human preferences. Safety training is one layer among...
- Sales Enablement
- The systems, content, training, and process support that help sales teams perform more effectively and consistently. Good sales enablement shortens ramp time...
- Scaling Law
- Scaling Law is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and...
- Self-Supervised
- Describing learning methods that use structure already present in unlabeled data to create training signals, rather than relying on externally labeled...
- Stochastic Gradient Descent
- Stochastic Gradient Descent is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning...
- Streaming Inference
- Streaming Inference is the phase where a trained model processes new inputs to produce predictions or generations. It is commonly used for production APIs,...
- Supervised
- Describing learning that uses labeled examples or explicit targets to teach a model the desired mapping from inputs to outputs. Supervised methods remain...
- Synthetic Data
- Artificially generated data that mimics the statistical properties of real data without containing actual personal or sensitive information. Used to augment...
- Synthetic Data
- Artificially generated data that mimics the statistical properties of real data, used when real data is scarce, expensive, or privacy-sensitive. The model...
- synthetic training data
- Training data generated by AI models rather than collected from humans, used to train or fine-tune other models. While dramatically cheaper to produce,...
- Teacher-Student Model
- A setup where a larger or stronger teacher model guides a smaller student model during training, often to transfer capability more efficiently.
- Tensor
- Tensor is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and research...
- Tensor Core
- Tensor Core is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and...
- TensorFlow
- TensorFlow is a machine learning framework for defining, training, and deploying numerical computation graphs. It is commonly used for deep learning research,...
- Tensor Parallelism
- Tensor Parallelism is a systems strategy for splitting model computation or parameters across devices. It is commonly used for training and serving workloads...
- TensorRT
- TensorRT is an NVIDIA inference optimization runtime for compiled neural networks. It is commonly used for serving GPU models with lower latency and higher...
- Test Set
- Test Set is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before release or model...
- TPU
- TPU is a specialized accelerator designed for large-scale tensor operations. It is commonly used for training and inference workloads with dense linear...
- Training Budget
- The amount of money, compute, time, or engineering effort available for training or fine-tuning a model. Training budgets constrain how much experimentation...
- Training Config
- The set of configuration values used for a training run, such as learning rate, batch size, optimizer choice, schedule, and data settings. Training config...
- Training Cost
- The total cost of training a model, including compute usage, storage, engineering time, data preparation, and supporting infrastructure. Training cost is a...
- Training Curriculum
- The planned ordering or progression of training examples, tasks, or difficulty levels over the course of learning. Training curricula are used when the...
- Training Distribution
- The distribution of examples, patterns, and conditions present in the data a model was trained on. A mismatch between training distribution and real-world use...
- Training Dynamics
- The behavior of the training process over time, including how loss, gradients, representations, and performance evolve during learning. Studying training...
- Training Efficiency
- How effectively a training setup converts compute, data, and time into improved model performance. Better training efficiency means reaching useful quality...
- Training Infrastructure
- The hardware, software, orchestration, storage, and operational systems used to run model training. Training infrastructure becomes a major engineering...
- Training Loss
- The loss measured on training data during model learning, used to indicate how well the model is fitting the current examples according to the objective....
- Training Loss Curve
- A graph showing how training loss changes over time, used to diagnose convergence, instability, or undertraining.
- Training Objective
- The objective function or goal that training is trying to optimize, such as next-token prediction, classification accuracy, or reward maximization. The...
- Training Pipeline
- The full pipeline that prepares data, configures jobs, runs training, evaluates checkpoints, and promotes successful artifacts. Training pipelines help make...
- Training Recipe
- A practical specification for how to train a model, including data setup, optimizer, schedules, regularization, and engineering tricks that together produce a...
- Training Schedule
- The plan governing how training progresses over time, such as learning rate schedules, phase transitions, checkpoint timing, or staged dataset exposure....
- Training Strategy
- The overall approach a team takes to training a model, including data choices, objectives, scaling decisions, adaptation methods, and evaluation criteria....
- Triplet Loss
- Triplet Loss is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for neural...
- Underfit
- To train a model that is too simple or insufficiently learned to capture the underlying pattern in the data.
- Upskilling
- The process of deliberately building new skills so someone can handle more advanced tools, roles, or responsibilities.
- Vanishing Gradient
- Vanishing Gradient is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for...
- Warm Start
- Starting training or optimization from an existing model, checkpoint, or good prior state instead of beginning from random initialization. Warm starts are...
- Weights
- The learned numerical parameters of a neural network, adjusted during training to minimize error. A model's weights are its accumulated knowledge — when...
Related Topics
- Ml (44 terms in common)
- Ai (38 terms in common)
- Optimization (35 terms in common)
- Evaluation (16 terms in common)
- Performance (8 terms in common)
- Systems (8 terms in common)
- Neural Networks (4 terms in common)
- Privacy (3 terms in common)
- Cost (3 terms in common)
- Learning (3 terms in common)
- Customization (2 terms in common)
- Generalization (2 terms in common)
- Labs (2 terms in common)
- Distillation (2 terms in common)
- Efficiency (2 terms in common)
- Normalization (2 terms in common)
- Alignment (2 terms in common)
- Safety (2 terms in common)
- Model (1 terms in common)
- Math (1 terms in common)