Special Sessions

New Trends in Continual Learning with Deep Architectures

Scope

One of the grand goals of Artificial Intelligence (AI) is building artificial “continual learning” agents that construct a sophisticated understanding of the world from their own experience through the autonomous development of increasingly complex knowledge and skills.

In the last few years, we have witnessed a rapidly growing interest in continual learning within the Machine Learning and Artificial Intelligence community, especially in the context of deep architectures for AI. However, current continual learning systems work in very limited scenarios and on oversimplified benchmarks. In particular, they are often designed and evaluated on standard computer vision multi-task supervised setting, which only slightly cover the entire space of possible scenarios a truly continual learning and autonomous agent may face.

In this special session proposal, we would like to focus our attention on new trends and revolutionary ideas for the future of this field: from new benchmarks and protocols to novel algorithms and methodologies and real-world deployment of continual learning systems.

Organizers

Vincenzo Lomonaco, University of Bologna, Italy

Davide Maltoni, University of Bologna, Italy

German I. Parisi, University of Hamburg, Germany

Topics

Topics of interest include, but are not limited to:

  • Benchmarks and Environments for Continual Learning
  • Evaluation Protocols and Metrics for Continual Learning
  • Continual Learning Strategies with Neural Networks
  • Alternative Learning Approaches for Continual Learning
    • Alternative Learning Rules
    • Bio-Inspired Approaches
    • Hybrid Systems
  • Computational and Memory Efficiency in Continual Learning
  • Continual Learning Applications
  • Multimodal Continual Learning
  • Online Continual Learning
  • Continual Learning in Multi-Task and Single-Incremental-Task Scenarios
  • Unsupervised and Semi-Supervised Continual Learning
  • Continual Reinforcement Learning
  • Sequence Learning and Active Learning in Continual Learning
  • Concept Drift and Distributional Shift Detection
  • Real-world Deployment Use-cases for Continual Learning Systems
  • Privacy, Safety and Security in Continual Learning

Supported by: