New geometric approach to learning in neural networks and quantum circuits

Researchers have developed a novel approach to machine learning by framing deep neural networks and variational quantum circuits as gradient flows on product Wasserstein manifolds. This geometric perspective treats weight distributions as intrinsic geometries rather than mere restrictions, potentially enhancing model capacity. The study introduces a hierarchical mean-field description for deep networks and extends it to quantum settings, proposing practical algorithms like Hierarchical DisCo-SGD and Quantum DisCo. Experiments indicate that this method improves generalization, stabilizes training, and mitigates barren plateaus in quantum neural networks. AI

IMPACT This geometric framing of learning could lead to more stable and generalizable models in both classical and quantum AI systems.

RANK_REASON Academic paper detailing a new theoretical framework and algorithms for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Deep Neural Networks
• Hierarchical DisCo-SGD
• Hierarchical mean-field description
• Quantum Wasserstein distance of order 1
• Variational Quantum Circuits
• Variational Quantum Classifiers
• Wasserstein Manifolds

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →