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.

Quaternion Extreme Learning Machine

Image super-resolution representation via image patches based on extreme learning machine

Extreme Learning Machine with Multiple Kernels

On fundamental aspects of quantum extreme learning machines

The augmented complex-valued extreme learning machine.

Quaternion-based machine learning on topological quantum systems

Experimental Results regarding multiple Machine Learning via Quaternions

A review on extreme learning machine

A Navel Facial Expression Recognition Method Based On Extreme Learning Machine

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

Dual Quaternion Based Virtual Hand Interaction Modeling

Extreme learning machine: algorithm, theory and applications

State estimation with quantum extreme learning machines beyond the scrambling time

Application of Quantum Extreme Learning Machines for QoS Prediction of Elevators' Software in an Industrial Context

Optimizing Extreme Learning Machine Via Generalized Hebbian Learning and Intrinsic Plasticity Learning

Affine phase retrieval of quaternion signals

The HR-Calculus: Enabling Information Processing with Quaternion Algebra

QDRJL: Quaternion dynamic representation with joint learning neural network for heart sound signal abnormal detection

Potential and limitations of quantum extreme learning machines

Learning Speech Emotion Representations in the Quaternion Domain