Researchers have developed "tnkm," an open-source Python library built with JAX for constructing and training Tensor Network Kernel Machines (TNKM). This framework aims to create nonlinear models that are both expressive and computationally efficient by combining nonlinear feature representations with compact low-rank tensor-network parameterizations. The library offers a unified interface for various feature maps, tensor-network architectures, and optimization strategies, including alternating least squares and gradient-based methods. Initial demonstrations on nonlinear benchmark problems show that TNKM models achieve competitive prediction accuracy with efficient training and compact parameterizations. AI
IMPACT This new framework could accelerate research and application of tensor-network-based learning methods, potentially leading to more efficient nonlinear models.
RANK_REASON The cluster describes a new open-source software library for machine learning research, detailed in an arXiv paper. [lever_c_demoted from research: ic=1 ai=1.0]
- Alternating Least-Squares for Low-Rank Matrix Reconstruction
- arXiv
- Gradient-based methods
- JAX
- Python
- Tensor Network Kernel Machines
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
