AI & Machine Learning Vocabulary

A computing system inspired by biological neural networks

“The neural network learned to recognize faces with remarkable accuracy.”

Origin: From Greek `neuron` (nerve) + Latin `rete` (net); coined in 1940s to describe brain-inspired computing

Machine learning using neural networks with many layers

“Deep learning revolutionized image recognition and natural language processing.”

Origin: Modern English compound; `deep` refers to multiple layers in neural networks

Training a model on labeled data with known outputs

“Supervised learning requires a dataset with correct answers for each example.”

Origin: From Latin `super` (over, above) + `videre` (to see), meaning oversight or guidance

Finding patterns in data without labeled examples

“Unsupervised learning discovered customer segments we hadn't considered.”

Origin: From Latin `un-` (not) + `super` (over) + `videre` (to see), meaning without guidance

Learning through trial and error with rewards and penalties

“Reinforcement learning enabled the AI to master complex games.”

Origin: From Latin `re-` (again) + `in-` (in) + `fortis` (strong), meaning to strengthen again

A neural network architecture using self-attention mechanisms

“Transformer models like GPT have transformed natural language processing.”

Origin: From Latin `trans-` (across) + `formare` (to form); transforms input sequences to output

A dense vector representation of data in continuous space

“Word embeddings capture semantic relationships between terms.”

Origin: From Old English `embeddan` (to fix firmly in a surrounding mass), from `em-` (in) + `bed`

Adapting a pre-trained model for a specific task

“Fine-tuning the base model on our data improved accuracy significantly.”

Origin: Modern English compound; `fine` (Old French `fin` meaning refined) + `tune` (Old English `tun` meaning tone)

Using a trained model to make predictions on new data

“Inference latency must be low for real-time applications.”

Origin: From Latin `inferre` (to bring in, conclude), from `in-` (in) + `ferre` (to carry)

When an AI generates false or fabricated information

“The model's hallucination produced a convincing but entirely fictional citation.”

Origin: From Latin `hallucinari` (to wander in mind, dream); metaphorical use for AI generating false information

Crafting inputs to elicit desired outputs from AI models

“Effective prompt engineering dramatically improved the response quality.”

Origin: From Latin `promptus` (brought forth) + Old French `engin` (skill), from Latin `ingenium` (cleverness)

Breaking text into smaller units for processing

“Tokenization splits sentences into words or subword pieces.”

Origin: From Old French `token` (sign, symbol), from Old English `tacn`; `-ization` from Greek suffix

A technique allowing models to focus on relevant parts of input

“The attention mechanism helps the model understand context across long sequences.”

Origin: From Latin `attendere` (to stretch toward, give heed), from `ad-` (to) + `tendere` (to stretch)

An optimization algorithm that minimizes error iteratively

“Gradient descent adjusts weights to reduce the loss function.”

Origin: From Latin `gradus` (step) + `descendere` (to go down), from `de-` (down) + `scandere` (to climb)

When a model learns noise instead of the underlying pattern

“Overfitting caused the model to perform poorly on new data.”

Origin: Modern English compound; `over` (excessive) + `fit` (to be suitable), meaning fitting too closely

When a model is too simple to capture the underlying pattern

“Underfitting resulted in poor performance on both training and test data.”

Origin: Modern English compound; `under` (insufficient) + `fit` (to be suitable), meaning not fitting closely enough

A parameter set before training begins, not learned from data

“Tuning hyperparameters like learning rate improved model performance.”

Origin: From Greek `hyper` (over, beyond) + `para` (beside) + `metron` (measure)

One complete pass through the entire training dataset

“The model converged after fifty epochs of training.”

Origin: From Greek `epokhe` (pause, fixed point in time), from `epi-` (upon) + `ekhein` (to hold)

The number of samples processed before updating the model

“Increasing batch size improved training stability but required more memory.”

Origin: From Old English `bæcce` (something baked) + `size` from Old French `sise` (portion)

A measure of how wrong the model's predictions are

“The loss function quantifies the difference between predictions and actual values.”

Origin: From Old English `los` (destruction, ruin) + Latin `functio` (performance), from `fungi` (to perform)

Prejudice in favor of or against one thing, person, or group

“We must ensure the AI model is free from bias.”

Origin: From French `biais` (slant, oblique), possibly from Greek `epikarsios` (athwart)

A process or set of rules to be followed in calculations

“The search algorithm ranks results by relevance.”

Origin: From Medieval Latin `algorismus`, from Arabic `al-Khwarizmi` (the mathematician from Khwarazm)