Abstract:Real-world 3D structured data like point clouds and skeletons often can be represented as data in a 3D rotation group (denoted as $\mathbb{SO}(3)$). However, most existing neural networks are tailored for the data in the Euclidean space, which makes the 3D rotation data not closed under their algebraic operations and leads to sub-optimal performance in 3D-related learning tasks. To resolve the issues caused by the above mismatching between data and model, we propose a novel non-real neuron model called \textit{quaternion product unit} (QPU) to represent data on 3D rotation groups. The proposed QPU leverages quaternion algebra and the law of the 3D rotation group, representing 3D rotation data as quaternions and merging them via a weighted chain of Hamilton products. We demonstrate that the QPU mathematically maintains the $\mathbb{SO}(3)$ structure of the 3D rotation data during the inference process and disentangles the 3D representations into ``rotation-invariant'' features and ``rotation-equivariant'' features, respectively. Moreover, we design a fast QPU to accelerate the computation of QPU. The fast QPU applies a tree-structured data indexing process, and accordingly, leverages the power of parallel computing, which reduces the computational complexity of QPU in a single thread from $\mathcal{O}(N)$ to $\mathcal {O}(\log N)$. Taking the fast QPU as a basic module, we develop a series of quaternion neural networks (QNNs), including quaternion multi-layer perceptron (QMLP), quaternion message passing (QMP), and so on. In addition, we make the QNNs compatible with conventional real-valued neural networks and applicable for both skeletons and point clouds. Experiments on synthetic and real-world 3D tasks show that the QNNs based on our fast QPUs are superior to state-of-the-art real-valued models, especially in the scenarios requiring the robustness to random rotations.

Experimental Results regarding multiple Machine Learning via Quaternions

Quaternion-based machine learning on topological quantum systems

Fast Quaternion Product Units for Learning Disentangled Representations in SO(3)

Exploring the Adjugate Matrix Approach to Quaternion Pose Extraction

Distance-Adaptive Quaternion Knowledge Graph Embedding with Bidirectional Rotation

Knowledge graph embedding with the special orthogonal group in quaternion space for link prediction

Quaternion Convolutional Neural Networks

Quaternion Representation Learning for cross-modal matching

A Survey on Dual-Quaternions

Provably Trainable Rotationally Equivariant Quantum Machine Learning

Learning Characteristics of Reverse Quaternion Neural Network

Learning Compact Transformation Based on Dual Quaternion for Point Cloud Registration

Quaternion-Based Self-Attentive Long Short-Term User Preference Encoding for Recommendation

Experimental Realization of a Quantum Support Vector Machine.

Augmented Quaternion and Augmented Unit Quaternion Optimization

Entanglement-Based Machine Learning On A Quantum Computer

Neural Networks at a Fraction with Pruned Quaternions

Dual quaternion QR decompositon and its corresponding complex structure-preserving algorithms

A Novel Robust Quaternions-Based Algorithm for 3-D Symmetric Similarity Datum Transformation

Randomized Large-Scale Quaternion Matrix Approximation: Practical Rangefinders and One-Pass Algorithm

Rotation-Equivariant Quaternion Neural Networks for 3D Point Cloud Processing