Research pills

Quaternion and Hypercomplex Neural Networks

A closer look to quaternion and hypercomplex neural networks, their advantages and the reasons behind their success

Quaternion Neural Networks

Hypercomplex Neural Networks

Paper references

[QNN1] T. Parcollet, M. Morchid, and G. Linarès, A survey of quaternion neural networks, in Artificial Intelligence Review, 53, pp. 2957–2982 (2020). [Paper]

[QNN2] E. Grassucci, E. Cicero, and D. Comminiello, Quaternion Generative Adversarial Networks, in Generative Adversarial Learning: Architectures and Applications, pp. 57-86 (2022). [Paper] [GitHub]

[PHNN1] E. Grassucci, A. Zhang, and D. Comminiello, PHNNs: Lightweight Neural Networks via Parameterized Hypercomplex Convolutions, arXiv preprint: arXiv:2110.04176 (2022). [Paper] [GitHub]

[PHNN2] A. Zhang, Y. Tay, S. Zhang, A. Chan, A. T. Luu, S. C. Hui, J. Fu, Beyond Fully-Connected Layers with Quaternions: Parameterization of Hypercomplex Multiplications with 1/n Parameters, in International Conference on Learning Representation (ICLR) (2021). [Paper]

Code references

HyperNets includes quaternion and parameterized hypercomplex layers in PyTorch with notebook tutorials.
Quaternion-Neural-Networks in PyTorch.