The Dictionary
Two corpora, one search. The vocabulary of AI and mathematics on one side, the working language of software engineering on the other — every term defined so it can be substituted back into a sentence, and every claim pointed at the paper it came from.
"exact phrase" · *ology wildcard ·
tensor NOT physics · misspellings are corrected automatically
Search
AI & Mathematics
The vocabulary of machine learning, statistics, optimisation and the linear algebra underneath all of it. Every entry is defined in terms a reader can substitute back into a sentence, then anchored to the paper or theorem it came from.
Activation Function
nouna mathematical function that introduces non-linearity into the output of a layer, allowing the network to learn complex patterns.
Attention Mechanism
nouna layer that produces each output as a weighted blend of all available inputs, where the weights are computed from the similarity between a query and a set of keys.
Backpropagation
nounan application of the chain rule that computes the sensitivity of a loss to every parameter in a layered network by passing partial derivatives from the output back toward the input.
Bayes' Theorem
nounthe identity that converts the likelihood of evidence given a hypothesis into the probability of the hypothesis given that evidence, by weighting it with the hypothesis's prior probability.
Bias-Variance Tradeoff
nounthe fundamental tension between the complexity of a model and its ability to generalize to unseen data.
Convolutional Neural Network
nounan architecture that slides a small set of learned filters across grid-structured input, so the same local pattern is recognised wherever it appears and far fewer weights are needed than a fully connected design.
Cross-Validation
nouna technique used to estimate the performance of a model on unseen data by partitioning the dataset into multiple subsets.
Eigenvector
nouna non-zero vector whose direction is unchanged by a given linear map, so the map merely rescales it by a constant factor.
Embedding
nouna low-dimensional vector representation of high-dimensional data, such as words or images, that captures semantic or structural relationships.
Entropy
nouna measure of the uncertainty or randomness inherent in a random variable or source of data.
Gradient Descent
nounan iterative procedure that reduces a differentiable objective by repeatedly stepping in the direction opposite to its slope at the current point.
Learning Rate
nouna hyperparameter that controls the step size taken at each iteration of an optimization algorithm.
Loss Function
nouna measure of the difference between predicted and actual values, quantifying the error of a model's output.
Markov Chain
nouna sequence of random variables where the probability of transitioning to the next state depends only on the current state, and not on the sequence of states that preceded it.
Maximum Likelihood Estimation
nouna method that picks the parameter values under which the observed data would have been most probable, by maximising the probability the model assigns to that data.
Overfitting
nounthe failure mode in which a model absorbs noise particular to its training sample and therefore predicts unseen data worse than a simpler model would.
Principal Component Analysis
nouna transformation that rewrites correlated measurements as a smaller set of uncorrelated axes, ordered so the first captures as much of the spread as any single direction can.
Regularization
nouna technique used to prevent overfitting by adding a penalty term to the loss function, which discourages overly complex models.
Reinforcement Learning
nouna training paradigm in which an agent improves its policy purely from rewards collected while acting in an environment, with no labelled examples of correct behaviour to imitate.
Softmax
nouna function that takes a vector of real numbers and maps it to a probability distribution, ensuring that all elements are positive and sum to one.
Transformer
nouna neural network architecture that relies solely on attention mechanisms, allowing it to process sequential data efficiently.
Software Engineering Core
The working vocabulary of building and running software — algorithms, architecture, concurrency, databases, security and the field's durable jargon. Definitions are written to be substitutable, and the idioms carry the origin story only where it is actually documented.
ACID
abbreviationthe guarantee that a group of statements either takes effect entirely or not at all, never observes a partial state, does not interfere with concurrent work, and survives a crash once confirmed.
Big O Notation
nouna way of describing how the cost of an algorithm grows as its input grows, keeping only the dominant term and discarding constants.
CAP Theorem
nounthe result that when a network splits, a replicated store must give up either answering every request or returning the newest written value — it cannot preserve both.
Cohesion
nounthe degree to which the parts of a single module all serve one purpose, rather than being grouped together for incidental reasons.
Coupling
nounthe degree to which one module depends on the internals of another, and therefore how far a change in one propagates into the other.
Deadlock
nouna state in which two or more processes are each waiting for the other to release a resource, resulting in a permanent halt of execution.
Dependency Injection
nounthe practice of supplying a component's collaborators from outside rather than letting it construct them, so the component can be reconfigured or tested without being modified.
Eventual Consistency
nounthe guarantee that if updates stop arriving, every replica will converge on the same values — with no promise about what a reader sees before then.
Hash Table
nouna container that converts each key into an array position by arithmetic, giving constant-time average lookup at the cost of handling positions that collide.
Idempotence
nounthe property of an operation whose repeated application produces the same result as a single application, so a retry cannot cause additional change.
Memoization
nounthe technique of storing the results of expensive function calls and returning the cached result when the same inputs occur again.
Mutex
nouna synchronization primitive used to ensure that only one thread can access a shared resource at any given time.
Pure Function
nouna routine whose result is determined entirely by its arguments and which changes nothing outside itself, so calling it twice with the same input is indistinguishable from calling it once.
Race Condition
nouna defect in which the correctness of a result depends on the unpredictable relative timing of two or more independent threads of execution.
Refactoring
nouna change to the internal structure of existing code that leaves its observable behaviour exactly as it was, undertaken to make later changes cheaper.
SQL Injection
nounan attack in which text supplied by a user is concatenated into a database query and thereby executed as part of the command rather than treated as a value.
Technical Debt
nounthe future cost incurred by choosing an expedient implementation now instead of a sounder one, paid back as extra effort on every later change.
Yak Shaving
nounthe state of working through a chain of unplanned prerequisite tasks, each apparently necessary for the last, so far from the original goal that the connection is no longer obvious.