Why This Tutorial?

Modern computer vision models depend on large, high-quality labeled datasets. But annotation is expensive, slow, and often requires domain expertise. Active Learning (AL) addresses this bottleneck by intelligently selecting the most informative and representative unlabeled samples for annotation.

This tutorial provides a comprehensive, hands-on introduction to active learning for computer vision. It combines theoretical foundations, practical strategies, evaluation methods, and a live coding session. By the end, attendees will understand how to build label-efficient computer vision systems that achieve strong performance with fewer annotations.

Overview

The performance of modern computer vision models is often constrained by the availability of labeled data. Active Learning offers a compelling solution: instead of labeling data randomly, an AL system iteratively selects the samples that are most useful for improving the model. This reduces annotation effort while maintaining or improving performance.

This tutorial is designed for graduate students, researchers, and practitioners who want to apply AL to real computer vision problems. It covers both classical AL strategies and modern approaches for deep learning, large-scale models, and foundation models.

What You Will Learn

By the end of this tutorial, participants will be able to:

• Explain the core paradigms of active learning: pool-based, stream-based, and membership query synthesis.

• Compare key query strategies, including uncertainty sampling, diversity sampling, query-by-committee, and hybrid methods.

• Understand the theoretical foundations of AL, including information theory, Bayesian optimization, and exploration–exploitation trade-offs.

• Implement a basic active learning pipeline in Python using PyTorch.

• Evaluate AL systems using classification and regression metrics.

• Adapt AL strategies for deep learning, Vision Transformers, CLIP-style models, and other foundation models.

• Identify practical applications in medical imaging, remote sensing, autonomous driving, and other domains.

Hands-On Session

The practical session will guide attendees through building an active learning pipeline from scratch.

Tools and dataset:

• Python

• PyTorch

• Jupyter Notebook

• Benchmark dataset such as CIFAR-10

Participants will:

1. Train an initial model on a small labeled subset.

2. Score unlabeled samples using an AL query strategy.

3. Select the most informative samples for annotation.

4. Simulate oracle labeling.

5. Retrain and evaluate the model.

6. Compare AL against random sampling.

Starter code and notebooks will be provided. Attendees are encouraged to bring a laptop, though the session will also be accessible as a guided demonstration.

Target Audience & Prerequisites

Target audience:

• Graduate students (MSc and PhD) + Postdoc and experts 

• Industry researchers

• Practitioners in computer vision and machine learning

Prerequisites:

• Basic understanding of machine learning and deep learning

• Familiarity with training neural networks, classification, and regression

• Working knowledge of Python and a deep learning framework such as PyTorch

• Basic probability and statistics

Expected attendance: 50–100 participants

Organizer

Dr. Alaa Tharwat

Dr. Alaa Tharwat is a Postdoctoral Researcher and Research Group Leader in the field of “Learning from Small Data” at the Bielefeld University of Applied Sciences and Arts, Bielefeld, Germany. His research focuses on data-centric AI, with a specific emphasis on Active Learning and its applications in computer vision, remote sensing, and medical imaging. He has published extensively in top-tier journals and conferences. His comprehensive survey, “A Survey on Active Learning: State-of-the-Art, Practical Challenges and Research Directions” (Tharwat & Schenck, 2023, Mathematics, 11(4), 820), received the Best Paper Award from the journal. He has organized multiple workshops and tutorials on data-efficient learning, active learning, and learning from small data at major international conferences, including ECML PKDD, IJCNN, and others.

Contact

For questions about the tutorial, please contact:

Dr. Alaa Tharwat
Bielefeld University of Applied Sciences and Arts, Germany
Email: alaa.othman@hsbi.de