Optimization Glossary
Browse 107 optimization terms defined in plain English, from the cultural dictionary of computing.
107 Optimization Terms
- 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 Efficiency
- The ability of an AI system to deliver useful results with minimal waste in compute, latency, memory, tokens, or operational overhead. AI efficiency matters...
- AI Optimization
- The process of improving an AI system's quality, latency, cost, or reliability through changes to prompts, models, retrieval, infrastructure, or evaluation. AI...
- AI Tuning
- The adjustment of prompts, parameters, retrieval settings, model choices, or fine-tuning configurations to improve AI system behavior. AI tuning is iterative...
- AutoML
- A collection of methods that automate model selection, feature engineering, and hyperparameter tuning. It influences how models are trained, evaluated, or...
- Backpropagation
- The algorithm that computes gradients through a network so weights can be updated efficiently. It influences how models are trained, evaluated, or served, and...
- 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,...
- Bayesian Optimization
- A sample-efficient optimization method that uses a surrogate model to choose promising hyperparameter trials. It influences how models are trained, evaluated,...
- Bit Twiddler
- A programmer who is skilled at or obsessed with low-level manipulation of bits, masks, shifts, and binary representations, often in embedded, systems, or...
- Cache
- Cache is a high-speed storage layer that keeps copies of frequently accessed data so future requests can be served faster than fetching from the original,...
- Code Performance
- How efficiently code runs in terms of speed, memory use, throughput, latency, and resource consumption under realistic conditions. Code performance matters not...
- Constraint Programming
- A programming approach in which problems are modeled as variables subject to constraints, and a solver searches for values that satisfy those constraints....
- Contrastive Loss
- A loss function that optimizes embeddings by rewarding similarity for matched pairs and separation for mismatched pairs. It influences how models are trained,...
- Copy-on-Write
- An optimization strategy where multiple references share the same data until one of them tries to modify it, at which point a copy is made for the modifier....
- Cross-Entropy Loss
- A loss function that measures how far predicted probability distributions are from the target distribution. It influences how models are trained, evaluated, or...
- 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,...
- 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...
- Direct Preference Optimization
- A post-training method that updates a model directly from paired preference data without an explicit reward model. It influences how models are trained,...
- 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,...
- DPO
- Abbreviation for direct preference optimization, a preference-based fine-tuning method for language models. It influences how models are trained, evaluated, or...
- 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,...
- Dynamic Programming
- An algorithm design technique that solves complex problems by breaking them into overlapping subproblems, solving each subproblem once, and storing the results...
- Dynamic Programming
- An optimization technique that solves problems by breaking them into overlapping subproblems and storing their solutions (memoization/tabulation) to avoid...
- Efficiency
- The ability to achieve useful results with relatively little wasted time, effort, or resources. In tech culture efficiency is often prized, though teams...
- 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,...
- 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...
- Federated Averaging
- A federated learning algorithm that aggregates model updates from many clients into a shared global model. It influences how models are trained, evaluated, or...
- Floating Point Quantization
- A compression technique that lowers numerical precision to reduce memory, bandwidth, and inference cost. It influences how models are trained, evaluated, or...
- Flow Matching
- A generative modeling method that learns continuous transport paths between simple and target distributions. It influences how models are trained, evaluated,...
- 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,...
- 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,...
- Greedy Algorithm
- An algorithmic paradigm that makes the locally optimal choice at each step, hoping to reach a globally optimal solution. Works for problems with the...
- Greedy Algorithm
- An algorithm that builds a solution incrementally by always making the locally optimal choice at each step, hoping this leads to a globally optimal solution....
- Hot Path
- The most frequently executed code path in a system — the sequence of operations that handles the majority of traffic or computation. Optimizing the hot path...
- 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,...
- Inference Optimization
- The process of making model inference faster, cheaper, or more efficient through techniques such as quantization, batching, caching, compilation, or smarter...
- 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,...
- INT4
- A four-bit integer numerical format commonly used for aggressive model quantization during inference. It influences how models are trained, evaluated, or...
- INT8
- An eight-bit integer numerical format used to reduce model size and speed up inference workloads. It influences how models are trained, evaluated, or served,...
- Intermediate Representation
- A data structure used internally by a compiler between the frontend (parsing) and backend (code generation) phases. Enables optimization passes that are...
- JIT Compiler
- A just-in-time compiler that translates code into native machine instructions during program execution instead of ahead of time. JITs watch running code,...
- Knowledge Distillation
- A model compression technique where a smaller student learns to mimic a larger teacher model. It influences how models are trained, evaluated, or served, and...
- lazy loading
- A technique that defers the loading of non-critical resources (images, scripts, components) until they're needed, typically when they enter or approach the...
- 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,...
- Learning to Rank
- A family of methods that train models to order results by relevance or utility. It influences how models are trained, evaluated, or served, and it can...
- LLM Optimization
- Improving a large language model workflow for quality, speed, reliability, or cost through changes to prompts, routing, context management, evaluation, or...
- Lookup Table
- A precomputed array or hash map that replaces expensive runtime calculations with direct indexed access, trading memory for speed. Common uses include CRC...
- Loop Unrolling
- A compiler or manual optimization that replicates the body of a loop multiple times per iteration, reducing the overhead of branch instructions, loop counter...
- Markov Decision Process
- Markov Decision Process is an evaluation concept used to measure model quality, robustness, or efficiency. It is commonly used for comparing systems before...
- Memoization
- An optimization technique that caches the results of expensive function calls and returns the cached result when the same inputs occur again. Only works for...
- Mesa-Optimization
- A concept in AI alignment referring to a learned subsystem that itself behaves like an optimizer pursuing objectives that may differ from the outer training...
- Model Distillation Loss
- The training loss used when a student model learns to match outputs or distributions from a teacher model during distillation.
- Model Optimization
- Improving a model or its serving path for better quality, efficiency, speed, or cost through tuning, pruning, quantization, routing, or infrastructure changes....
- Multivariate Testing
- An experimentation method that tests combinations of multiple variable changes at the same time to understand both individual and interaction effects. It is...
- Neural Network Pruning
- The removal of weights, neurons, or connections from a neural network to make it smaller or more efficient while trying to preserve performance. Pruning is one...
- 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...
- Offline Reinforcement Learning
- Offline Reinforcement Learning is a learning paradigm that improves task performance from data, feedback, or experience. It is commonly used for models that...
- Penalty
- A cost or discouraging adjustment applied during optimization or generation to reduce unwanted behavior such as repetition, excessive length, or policy...
- Policy Gradient
- Policy Gradient is a training-time optimization concept that governs how model parameters are updated. It is commonly used for iterative learning loops for...
- PPO
- PPO is a policy-gradient reinforcement learning algorithm that constrains policy updates with clipping. It is commonly used for fine-tuning agents and language...
- PPO Algorithm
- Proximal Policy Optimization, a reinforcement-learning algorithm often used in policy training and historically common in RLHF pipelines.
- Prefetch
- To speculatively load data, assets, or DNS resolutions before they are explicitly needed, anticipating that the user or program will request them soon. In...
- Production Build
- A compiled and optimized version of an application intended for end users, with minification, tree-shaking, dead code elimination, source map separation, and...
- Profile
- To measure where a program spends its time and memory, identifying bottlenecks. Profilers instrument code to collect metrics — CPU time, memory allocations,...
- Profile Guided Optimization
- A two-pass compiler optimization technique where the program is first compiled with instrumentation, then run against representative workloads to collect...
- Profiler
- A tool that measures where a program spends its time and memory, typically by sampling the call stack at intervals or instrumenting function entry/exit....
- Prompt Optimization
- Improving prompt performance for quality, consistency, cost, or latency through iteration, testing, and measurement. Prompt optimization often yields large...
- Proximal Policy Optimization
- Proximal Policy Optimization is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production...
- Q-Learning
- Q-Learning is a learning paradigm that improves task performance from data, feedback, or experience. It is commonly used for models that adapt representations...
- Quantization
- Reducing the precision of a model's numerical weights (e.g., from 32-bit to 4-bit) to decrease memory usage and increase inference speed, with minimal quality...
- query planner
- The database component that analyzes a SQL query and determines the most efficient execution strategy — which indexes to use, what join order and algorithms to...
- Reinforcement Learning Detail
- Reinforcement Learning Detail is a learning paradigm that improves task performance from data, feedback, or experience. It is commonly used for models that...
- Reinforcement Learning from AI Feedback
- Reinforcement Learning from AI Feedback is a learning paradigm that improves task performance from data, feedback, or experience. It is commonly used for...
- Reward Engineering
- The design and tuning of reward signals, reward models, or scoring functions so learning systems optimize for the intended behavior. Reward engineering is...
- Reward Function
- Reward Function is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and...
- Reward Hacking
- Reward Hacking is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models and...
- Reward Model
- Reward Model is a model component or design choice that shapes how information flows through a learned system. It is commonly used for building neural...
- Score Function
- A function that assigns scores to outputs, actions, candidates, or states so they can be compared, ranked, or optimized. Score functions appear in ranking,...
- StormForge
- A tooling brand associated with optimizing Kubernetes resource settings and performance behavior.
- String Pool
- A memory region (most notably in the JVM) where string literals are interned so that identical string values share a single object reference, reducing heap...
- Support Vector Machine
- Support Vector Machine is an AI or ML concept used to represent, train, evaluate, or deploy learned systems. It is commonly used for building production models...
- Surrogate Model
- A simpler or cheaper model used to approximate a more expensive process, objective, or system during optimization or experimentation. Surrogate models help...
- Tail Call
- A function call that occurs as the last action of a function, allowing the compiler to reuse the current stack frame instead of allocating a new one. Enables...
- Time to First Byte Culture
- A performance-minded culture where teams care deeply about responsiveness metrics such as time to first byte and treat latency as a product and infrastructure...
- Training Dynamics
- The behavior of the training process over time, including how loss, gradients, representations, and performance evolve during learning. Studying training...
- Training Loss Curve
- A graph showing how training loss changes over time, used to diagnose convergence, instability, or undertraining.
- Tree Shake Slang
- Informal use of tree shake for removing unused code or dependencies so only necessary pieces remain. In engineering slang, people sometimes use it more broadly...
- Unrolling
- A compiler or manual optimization technique that replaces a loop with repeated copies of its body, reducing the overhead of branch instructions and...
- Vectorization
- Transforming scalar operations into SIMD (Single Instruction, Multiple Data) operations that process multiple data elements in a single CPU instruction. Can...
Related Topics
- Training (35 terms in common)
- Ml (26 terms in common)
- Ai (18 terms in common)
- Performance (16 terms in common)
- Reinforcement Learning (13 terms in common)
- Agents (12 terms in common)
- Inference (4 terms in common)
- Memory (3 terms in common)
- Algorithms (3 terms in common)
- Alignment (3 terms in common)
- Quantization (3 terms in common)
- Compilers (3 terms in common)
- Runtime (2 terms in common)
- Functional (2 terms in common)
- Neural Networks (2 terms in common)
- Experimentation (2 terms in common)
- Compression (2 terms in common)
- Deployment (2 terms in common)
- Debugging (2 terms in common)
- Models (2 terms in common)