AI 용어집
인공지능 완전 사전
Self-Paced Learning
Variant of curriculum learning where the model learns at its own pace by dynamically selecting training examples based on their perceived difficulty level.
Difficulty Scoring
Quantitative method to evaluate and assign a difficulty score to each training example, allowing for optimal scheduling in the curriculum.
Baby Steps Learning
Extreme curricular approach where the model starts with trivially simple examples before progressing very gradually towards complex cases.
MentorNet
Meta-learning neural network that dynamically defines the curriculum by learning to select the most relevant examples for the student network.
Curriculum by Data Density
Strategy that orders examples according to their density in the feature space, prioritizing samples in dense regions before those in sparse regions.
Curriculum by Gradient Noise
Method that uses noise in gradients as a difficulty indicator, with examples having more noise being considered more difficult and introduced later.
Curriculum by Loss
Approach that orders examples according to their current loss value, with examples with high loss being considered difficult and delayed in the curriculum.
Anti-Curriculum Learning
Inverse strategy that presents the most difficult examples first to improve model robustness and avoid suboptimal local minima.
Curriculum by Transfer Learning
Method using pre-learned knowledge on simple tasks to build a progressive curriculum towards more complex tasks.
Curriculum Generation
Algorithmic process of automatically creating optimal learning sequences based on data characteristics and model objectives.
Curriculum Scheduling
Temporal planning defining when and how to introduce examples of increasing difficulty during model training.
Easy-First Strategy
Fundamental principle of curriculum learning where the simplest examples are presented first to establish a solid foundation before complexity.
Hard-First Strategy
Alternative approach presenting difficult examples first to force the model to develop robust representations from the beginning of learning.
Curriculum Smoothness
Measure of continuity in the difficulty progression between successive examples, avoiding abrupt jumps that could destabilize learning.
Task Curriculum Learning
Extension of curriculum learning where the order applies to entire tasks rather than individual examples within the same task.
Multi-Task Curriculum Learning
Complex approach orchestrating simultaneous learning of multiple tasks with an optimized curriculum to maximize synergies between tasks.