The goal of this special session is to leverage the strengths of multimodal foundation models and evolutionary computation to address the challenges of building autonomous agents that can perceive, learn, and adapt in dynamic, real-world environments.
Topics of interest include, but are not limited to:
Evolving and lifelong-learning models for multimodal agent adaptation and the mitigation of catastrophic forgetting.
Unified vision-language-audio architectures and cross-modal grounding techniques for robust situational understanding.
Neuro-evolutionary approaches and genetic programming strategies for evolving autonomous agent policies.
Multimodal planning and hierarchical reasoning frameworks for complex goal decomposition and error recovery.
Advanced memory systems (episodic, semantic, and working memory) for agents operating in long-context scenarios.
Multi-agent coordination and emergent communication in heterogeneous multimodal settings.
Safety, value alignment, and explainability mechanisms for monitoring and auditing agents as they evolve.
Efficient and edge deployment strategies, including quantization, split inference, and privacy-preserving multimodal processing.
Self-supervised and meta-learning techniques for agent generalization in non-stationary and resource-constrained environments.
Benchmarks, metrics, and human-in-the-loop evaluation protocols for measuring evolutionary capabilities and adaptation quality.
Real-world applications and case studies of adaptive multimodal agents in Healthcare, Industry 5.0, Robotics, and Smart Environments.