sellThesis sellDFKI sellAugmented Vision sellDeep Learning

Classroom-Inspired Learning for Deep Neural Networks (Thesis)

Shalini Sarode , Muhammad Saif Ullah Khan

Abstract

This thesis investigates classroom-inspired learning strategies for training deep neural networks. The central idea is to structure supervision, pacing, and sample difficulty similarly to pedagogical curricula, with the objective of improving convergence stability and generalization.

Topic

Design and evaluate training strategies inspired by classroom learning, including curriculum ordering, adaptive difficulty, and staged supervision. Compare against standard training schedules on human-centric vision tasks.

Tasks

1. Implement curriculum-based training strategies in a PyTorch pipeline.
2. Define measurable learning schedules (sample difficulty, pacing, and progression criteria).
3. Evaluate effects on convergence, stability, and generalization.

Related Literature

• [1] Curriculum Learning
• [2] Self-Paced Learning for Latent Variable Models
• [3] MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks

Maintained by saifkhichi96 on GitHub.

The website is distributed under different open-source licenses. For more details, see the notice at the bottom of the page.