About the Workshop

Recent advancements in AI are largely perceived as monolithic systems, built around a single model trained on a vast dataset. Despite their remarkable results, businesses encounter significant challenges with these systems, including resource-intensive updates and difficulty recruiting AI talent.

This workshop advocates for exploring modern paradigms that emphasize rapid adaptability, model customization, and data heterogeneity through Modular Deep Learning. This emerging framework enables the development of AI models composed of efficient, interdependent modules, each specialized for specific purposes and designed for consistent reusability.

We aim to foster research on underexplored aspects of Modular Deep Learning, with a focus on its role in Continual Learning, Federated Learning, and generative models, bridging the gap between theoretical advancements and real-world applications.

Topics include but are not limited to:

• Modular approaches for Continual Learning and Federated Learning
• Model merging techniques and Mixture of Experts strategies
• Innovative methodologies for modular model editing and patching
• Development of libraries and frameworks for modular learning
• Evaluation metrics for modular AI systems
• Case studies of Modular Deep Learning in various domains