Speaks the AI language

Use this glossary to express differences, not just memorize acronyms. Each group links to the module where you can apply them.

Data and Learning

Term	Practical Meaning
AI / AI	Field systems performing perception, language or decision tasks.
ML	Learning patterns from data.
Feature	Input feature available while predicting.
Label / target	Answer known to train a supervised model.
Recovery	A prediction of a numerical value.
Classification	A category assignment.
Clustering	Grouping without target labels.
Inference	Application from a model trained to an entrance.
Loss	A measure of the error that guides training.
Overfitting	That's good for training and bad generalization.
Data leakage	Inadequate information about the target or test that contaminates training.
Precision / recall	Proportion of right alerts / ratio of actual positive found.
Policy, RL	Strategy to select actions under state.

Term	Practical Meaning
Weekly / bias	Weight of a connection / term that moves a sum.
Activation	Transformation applied to the weighted sum of a neuron.
Backspread	Gradient calculation through layers during training.
CNN	Network that learns filters about local patterns.
RNN / LSTM	state sequence networks LSTM regulates memory with gates.
Transformer	Architecture that uses care to relate positions.
Token	Unit of the tokenizer That can be part of a word.
Parameter	Value learned from the model That's not an entrance token.
Context window	Context capacity of an invocation.
Embedding	Vector that represents relationships learned from the content.
Faw-shot	Examples within the prompt without an update of weights.
Training tuning	Training to follow instructions.
Fine-tuning	To adapt cohesive parameters with specific examples.
RAG / ground	Recovery and generation with context / support of statements in evidence.
Fallucination	Wrong or unsupported content submitted as a response.

Term	Practical Meaning
Model Catalog	Data Science model and metadata repository.
Model Deploment	Resource that serves an inference model.
Job	Reducible performance, such as training or evaluation.
Playground	Test model interface, prompts and parameters.
Dedicated AI Cluster	dedicated photo for fine-tuning or hosting according to resource.
GPU / RDMA	Processing parallel / memory transfer with low CPU intervention.
ASR / SRT	Automatic voice recognition / subtitle format.
Diarization	Separation of segments by speaker without proving his civil identity.
OCR	Retirement of text in images or documents.
NER	Extraction of designated entities from text.
Bounding box	A box that locates an object or fragment detected.
pII / phi	Identifiable personal information / protected health information.

Relationship to each term with the portfolio y the services of AI.

Term	Practical Meaning
Tool / function calling	Operation offered to agent / mechanism to request his execution.
Runtime / loop	The environment that implements and coordinates / decision cycle, action and observation.
Short-term / long-term memory	Context from a conserved conversation / information between conversations.
MCP	Connecting protocol between applications and tool or context servers.
Authentication / authorization	Test identity / check permissions.
Guarrail	Control of entry or exit risks supplementing the authorisation.
VECTOR / vector index	Type for embeddings / structure that accelerates your search.
Select AI / Select AI Agent	Natural language interaction with data and co-ordination of agents and tools.

Reg OCI Enterprise AI e AI for databases. Test explain why memory, training and authorization are three different things.