Domain: Foundations & Theory

127 terms

Learning Rate Intermediate

Controls the size of parameter updates; too high diverges, too low trains slowly or gets stuck.

LIME Intermediate

Local surrogate explanation method approximating model behavior near a specific input.

Line Search Intermediate

Choosing step size along gradient direction.

Linear Quadratic Regulator Intermediate

Optimal control for linear systems with quadratic cost.

Local Minimum Intermediate

Minimum relative to nearby points.

Logits Intermediate

Raw model outputs before converting to probabilities; manipulated during decoding and calibration.

LoRA Intermediate

PEFT method injecting trainable low-rank matrices into layers, enabling efficient fine-tuning.

Loss Function Intermediate

A function measuring prediction error (and sometimes calibration), guiding gradient-based optimization.

LSTM Intermediate

An RNN variant using gates to mitigate vanishing gradients and capture longer context.

Lyapunov Stability Intermediate

Stability proven via monotonic decrease of Lyapunov function.

Masked Language Model Intermediate

Predicts masked tokens in a sequence, enabling bidirectional context; often used for embeddings rather than generation.

Memory Intermediate

Mechanisms for retaining context across turns/sessions: scratchpads, vector memories, structured stores.

Model Intermediate

A parameterized mapping from inputs to outputs; includes architecture + learned parameters.

Model Card Intermediate

Standardized documentation describing intended use, performance, limitations, data, and ethical considerations.

Model Predictive Control Intermediate

Optimizes future actions using a model of dynamics.

Model Registry Intermediate

Central system to store model versions, metadata, approvals, and deployment state.

Model Stealing Intermediate

Reconstructing a model or its capabilities via API queries or leaked artifacts.

Multimodal Model Intermediate

Models that process or generate multiple modalities, enabling vision-language tasks, speech, video understanding, etc.

Next-Token Prediction Intermediate

Training objective where the model predicts the next token given previous tokens (causal modeling).

NLP Intermediate

AI subfield dealing with understanding and generating human language, including syntax, semantics, and pragmatics.

Normalization Intermediate

Techniques that stabilize and speed training by normalizing activations; LayerNorm is common in Transformers.

Objective Surface Intermediate

Visualization of optimization landscape.

Optimal Control Intermediate

Finding control policies minimizing cumulative cost.

Orchestration Intermediate

Coordinating tools, models, and steps (retrieval, calls, validation) to deliver reliable end-to-end behavior.

Overfitting Intermediate

When a model fits noise/idiosyncrasies of training data and performs poorly on unseen data.

Parameter-Efficient Fine-Tuning Intermediate

Techniques that fine-tune small additional components rather than all weights to reduce compute and storage.

Parameters Intermediate

The learned numeric values of a model adjusted during training to minimize a loss function.

PID Controller Intermediate

Classical controller balancing responsiveness and stability.

PII Intermediate

Information that can identify an individual (directly or indirectly); requires careful handling and compliance.

Planning Intermediate

Methods for breaking goals into steps; can be classical (A*, STRIPS) or LLM-driven with tool calls.