ChemicalQDeviceTensor Network, Neural Network, or Hybrid
Tensor Network, Neural Network, or Hybrid AI Practical Use 03-14-24.pdf

Next level AI: Tensor Networks and Neural Networks are addressing the Curse of Dimensionality in significant ways. A 2023 Wang, M., et al. paper titled "Tensor Networks Meet Neural Networks: A Survey and Future Perspectives" features much research on TNs being used for Network compression, information fusion, and quantum circuit simulation. Works include TN utilization with popular machine learning models such as Transformers, CNNs, RNNs, GNNs, RBMs, FLs, Quantum Embedding, and Quantum TNNs.

In addition, Neural network and tensor network hybrid models have been created to utilize benefits of both network types for expressivity and an inductive bias. The NSF/MIT/Harvard 'ANTN' model improves on two 2022 papers that were not able to address correct wavefunction behavior properly. The neural network and incorporated MPS tensor network now contain a necessary physics prior to efficiently represent local or quasi-local sign structure, in a unique quantum-inspired 'pretraining' manner.

JuliaCon in 2021, 2022, and 2023 has featured developments from Flatiron Institute's ITensors.jl which is a library for rapidly creating correct and efficient tensor network algorithms, and has been used in many hundreds of papers. ITensorNetworks.jl extends on ITensors.jl and provides general network data structures and tools. ITensorNetworks may be a key library for advancing the pace of AI into the future for high dimensional problems by using Tensor network solvers accessible to more general classes of problems.

Created by Kevin Kawchak Founder CEO ChemicalQDevice2024 San Diego, CaliforniaHealthcare Innovation