Automated Architecture Search for Brain-inspired Hyperdimensional Computing

Junhuan Yang, Yi Sheng, Sizhe Zhang, Ruixuan Wang, Kenneth Foreman, Mikell Paige, Dayane Reis, Xun Jiao, Weiwen Jiang, Lei Yang

Abstract: This paper represents the first effort to develop an automated architecture search framework for hyperdimensional computing (HDC), a type of brain-inspired neural network. The framework, named AutoHDC, fills the gap in the optimization of HDC architecture design for given applications, which is currently carried out in an application-specific ad-hoc manner. Automated exploration will not only push HDC to more general applications, but also significantly diminish the heavy labor in architecture optimization for high performance and efficiency. To enable automated exploration, we present a thorough study to formulate the HDC architecture search space. On top of this, we apply reinforcement learning to automatically explore the HDC architectures. AutoHDC is evaluated in case studies on drug screening tasks in drug discovery. On the ClinTox dataset, AutoHDC can identify an architecture that outperforms the state-of-the-art deep learning approach with 4.77% higher ROC-AUC scores on average, and 2.26% higher scores against the manually designed HDC.

Keywords: Hyperdimensional Computing, Automated HDC, Automated Architecture Search

One-sentence Summary: AutoHDC conduct an automated architecture search for hyperdimensional computing.

Reproducibility Checklist: Yes

Broader Impact Statement: Yes

Paper Availability And License: Yes

Code Of Conduct: Yes

Reviewers: <Weiwen Jiang>,<[e-mail]>