P (44 terms)
PAC Learning
A model is PAC-learnable if it can, with high probability, learn an approximately correct hypothesis from finite samp...
Intermediate
Parameter Sharing
Using same parameters across different parts of a model.
Intermediate
Parameter-Efficient Fine-Tuning
Techniques that fine-tune small additional components rather than all weights to reduce compute and storage.
Intermediate
Parameters
The learned numeric values of a model adjusted during training to minimize a loss function.
Intermediate
Pareto Optimality
No agent can improve without hurting another.
Advanced
Particle Filter
Monte Carlo method for state estimation.
Intermediate
Path Planning
Finding routes from start to goal.
Advanced
Perception Stack
Software pipeline converting raw sensor data into structured representations.
Advanced
Perplexity
Exponential of average negative log-likelihood; lower means better predictive fit, not necessarily better utility.
Intermediate
Physical Safety
Ensuring robots do not harm humans.
Frontier
Physics Engine
Software simulating physical laws.
Advanced
PID Controller
Classical controller balancing responsiveness and stability.
Intermediate
PII
Information that can identify an individual (directly or indirectly); requires careful handling and compliance.
Intermediate
Planner-Executor
Separates planning from execution in agent architectures.
Intermediate
Planning
Methods for breaking goals into steps; can be classical (A*, STRIPS) or LLM-driven with tool calls.
Intermediate
Planning Horizon
Number of steps considered in planning.
Advanced
Plant
The physical system being controlled.
Intermediate
Polarization
Groups adopting extreme positions.
Advanced
Policy
Strategy mapping states to actions.
Intermediate
Policy Gradient
Optimizing policies directly via gradient ascent on expected reward.
Intermediate
Policy Search
Directly optimizing control policies.
Advanced
Positional Encoding
Injects sequence order into Transformers, since attention alone is permutation-invariant.
Intermediate
Posterior Distribution
Updated belief after observing data.
Advanced
Potential Fields
Planning via artificial force fields.
Advanced
Power-Seeking Behavior
Tendency to gain control/resources.
Advanced
PR Curve
Often more informative than ROC on imbalanced datasets; focuses on positive class performance.
Intermediate
Precision
Of predicted positives, the fraction that are truly positive; sensitive to false positives.
Intermediate
Prediction Drift
Shift in model outputs.
Intermediate
Predictive Coding
Learning by minimizing prediction error.
Frontier
Predictive Policing
AI predicting crime patterns (highly controversial).
Intermediate
Prior Distribution
Belief before observing data.
Advanced
Privacy Attack
Attacks that infer whether specific records were in training data, or reconstruct sensitive training examples.
Intermediate
Probability Distribution
Describes likelihoods of random variable outcomes.
Advanced
Prognostic Model
Predicting disease progression or survival.
Intermediate
Prompt
The text (and possibly other modalities) given to an LLM to condition its output behavior.
Intermediate
Prompt Engineering
Crafting prompts to elicit desired behavior, often using role, structure, constraints, and examples.
Intermediate
Prompt Injection
Attacks that manipulate model instructions (especially via retrieved content) to override system goals or exfiltrate ...
Intermediate
Prompt Leakage
Extracting system prompts or hidden instructions.
Intermediate
Prompt Sensitivity
Small prompt changes cause large output changes.
Intermediate
Propensity Score
Probability of treatment assignment given covariates.
Advanced
Proprioception
Internal sensing of joint positions, velocities, and forces.
Advanced
Prosody
Temporal and pitch characteristics of speech.
Intermediate
Protein Folding
Predicting protein 3D structure from sequence.
Advanced
Pruning
Removing weights or neurons to shrink models and improve efficiency; can be structured or unstructured.
Intermediate