Curriculum Learning: An Efficient Learning Paradigm

Curriculum Learning (CL) is a brain-inspired training paradigm that mirrors how humans and animals acquire complex skills—by learning from simple concepts first and gradually progressing to harder ones. Although CL has been shown to improve convergence speed, optimization stability, generalization, and computational efficiency, it remains significantly under-explored within the Sri Lankan machine learning research community.

An analysis of nearly 2,000 locally published research papers reveals that most studies focus on conventional paradigms such as supervised and unsupervised learning, while very few investigate Curriculum Learning or related strategies. This gap presents a unique opportunity for researchers and practitioners to contribute novel, high-impact work in an area with strong theoretical grounding, broad applicability, and growing relevance to modern AI systems.

This workshop is motivated by the need to bridge this gap by providing both conceptual clarity and practical exposure to Curriculum Learning, enabling participants to understand why it works, when it is effective, and how it can be systematically designed and applied in real-world deep learning pipelines.

The primary objective of this workshop is to equip participants with a strong foundation in Curriculum Learning, covering both theoretical principles and hands-on implementation strategies. By the end of the workshop, participants will be able to:

• Understand the intuition and theoretical foundations behind Curriculum Learning.
• Design and analyze curriculum strategies using scoring functions and pacing schedules.
• Compare baseline training, curriculum learning, and anti-curriculum regimes.
• Evaluate the impact of curricula on training stability, convergence behavior, and generalization.
• Apply Curriculum Learning across domains such as computer vision, natural language processing, and multi-modal learning.

• Foundations of Curriculum Learning
  Brain-inspired learning, continuation methods, and the role of structured difficulty in optimization.
• Difficulty Modeling
  How to define and measure sample difficulty using data-level and model-level scoring functions.
• Training Schedulers (Pacing Functions)
  Strategies for controlling when and how harder samples are introduced during training.
• Curriculum Variants
  Self-paced learning, teacher–student curricula, adaptive curricula, and comparisons with anti-curriculum learning.

Curriculum Learning is presented as an architecture-agnostic and paradigm-independent strategy applicable across:

• Computer Vision – image classification and structured visual understanding.
• Natural Language Processing – learning from simple linguistic structures to complex semantics.
• Graph Neural Networks – integrating curricula across graph data.

Hands-on experiments will demonstrate how curriculum design influences learning behavior in deep networks. Participants will work with reproducible implementations that illustrate efficiency gains, robustness improvements, and behavioral differences across training regimes.