FingHV: Efficient Sharing and Fine-Grained Scheduling of Virtualized HPU Resources

Hui Wang , Zhiwen Yu , Zhuoli Ren

IEEE Transactions on Cybernetics +

|

March 3, 2025

View abstract on PubMed

Summary

This study introduces fine-grained human processing unit (HPU) virtualization for human-machine computing (HMC) systems. The new approach enhances task assignment and resource utility in collaborative AI environments.

Area Of Science

Computer Science
Artificial Intelligence
Human-Computer Interaction

Background

The advancement of artificial intelligence (AI) necessitates human-centered design and collaborative systems.
Human-machine computing (HMC) paradigms combine human cognition with machine computation for complex tasks.
Efficient resource provisioning for Human Processing Units (HPUs) is critical for HMC system performance.

Purpose Of The Study

To address limitations in existing HPU resource provisioning schemes that fail to optimize task assignment and utility.
To propose a novel fine-grained HPU virtualization (FingHV) approach for improved flexibility, fairness, and utility.
To enhance human-machine symbiosis in large-scale, complex task environments.

Main Methods

Development of a hierarchical multiskill tree to model HPU skills and their correlations.
Formulation of the HPU virtualization problem.
Implementation of a fine-grained virtualization method including quality-driven HPU assignment and mixed time/event-based scheduling.

Main Results

The proposed FingHV approach significantly improves global matching quality by up to 39.7%.
FingHV increases HPU utility by 11.2% compared to existing baseline methods.
Evaluation on synthetic and real-world datasets validates the effectiveness of the FingHV approach.

Conclusions

Fine-grained HPU virtualization is essential for optimizing resource allocation in HMC systems.
The FingHV approach offers a robust solution for enhancing collaborative AI performance.
This work advances the field of human-machine symbiosis through improved resource management.