Abstract:Twisted Convolutional Networks (TCNs) are introduced as a novel neural network architecture designed to effectively process one-dimensional data with arbitrary feature order and minimal spatial relationships. Unlike traditional Convolutional Neural Networks (CNNs), which excel at handling structured two-dimensional data like images, TCNs reduce dependency on feature order by combining input features in innovative ways to create new representations. By explicitly enhancing feature interactions and employing diverse feature combinations, TCNs generate richer and more informative representations, making them especially effective for classification tasks on datasets with arbitrary feature arrangements. This paper details the TCN architecture and its feature combination strategy, providing a comprehensive comparison with traditional CNNs, DeepSets, Transformers, and Graph Neural Networks (GNNs). Extensive experiments on benchmark datasets demonstrate that TCNs achieve superior performance, particularly in classification scenarios involving one-dimensional data.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the limitations of traditional convolutional neural networks (CNNs) when processing non - spatial data. Specifically, CNNs rely heavily on the spatial order of input features, which makes them perform poorly when dealing with data without a clear spatial structure or feature relationships. For example, in application scenarios such as gene expression data, customer demographic information, and sensor readings, the relationships between features are not strictly spatial or sequential, and the order of feature arrangement may not carry important information. To overcome these limitations, the paper proposes a new neural network architecture - Twisted Convolutional Networks (TCNs). The main goal of TCNs is to combine input features in an innovative way, reduce the dependence on feature order, and generate richer and more informative feature representations. This makes TCNs particularly effective when dealing with data sets with arbitrary feature order, especially in classification tasks. ### Main problem summary: 1. **Limitations of traditional CNNs**: CNNs depend on the spatial order of features and are not effective for non - spatial data (such as one - dimensional data). 2. **Importance of feature combination**: Existing methods fail to fully utilize the potential information in feature combinations. 3. **Improving model robustness**: By enhancing feature interaction, the model is made more robust when dealing with data with arbitrary feature arrangements. ### Solutions: - **Introducing TCNs**: Combine input features in multiple ways to generate new feature representations, thereby reducing the dependence on feature order. - **Feature combination strategies**: Include two methods, multiplicative combination and pairwise product summation, to capture high - order interactions between features. - **Improved network architecture**: Include a feature interaction module, residual connections, and regularization techniques to improve the generalization ability and training stability of the model. Through these improvements, TCNs can achieve better classification performance on various benchmark data sets, especially in classification scenarios involving one - dimensional data.

Twisted Convolutional Networks (TCNs): Enhancing Feature Interactions for Non-Spatial Data Classification

A Novel Transformer Network with a CNN-Enhanced Cross-Attention Mechanism for Hyperspectral Image Classification

A Tensor-based Convolutional Neural Network for Small Dataset Classification

Transform-Invariant Convolutional Neural Networks for Image Classification and Search

An Alternative Practice of Tropical Convolution to Traditional Convolutional Neural Networks

3DCTN: 3D Convolution-Transformer Network for Point Cloud Classification

Hypergraph-Induced Convolutional Networks for Visual Classification

Tensorial Neural Networks: Generalization of Neural Networks and Application to Model Compression

Convolution Transformer Fusion Splicing Network for Hyperspectral Image Classification

Cross-GCN: Enhancing Graph Convolutional Network with K-Order Feature Interactions

STCN: Stochastic Temporal Convolutional Networks

TCCT: Tightly-coupled convolutional transformer on time series forecasting

Adaptive Gabor convolutional networks

Graph Convolutional Networks for Hyperspectral Image Classification

CNN and Transformer interaction network for hyperspectral image classification

GTFN: GCN and Transformer Fusion Network With Spatial-Spectral Features for Hyperspectral Image Classification

Convolutional Neural Networks Exploiting Attributes of Biological Neurons

TSC: A Simple Two-Sided Constraint against Over-Smoothing

Composite Neighbor-Aware Convolutional Metric Networks for Hyperspectral Image Classification

Temporal Convolutional Explorer Helps Understand 1D-CNN's Learning Behavior in Time Series Classification from Frequency Domain

A Joint Convolutional Cross ViT Network for Hyperspectral and Light Detection and Ranging Fusion Classification