Abstract:The intrinsically infinite-dimensional features of the functional observations over multidimensional domains render the standard classification methods effectively inapplicable. To address this problem, we introduce a novel multiclass functional deep neural network (mfDNN) classifier as an innovative data mining and classification tool. Specifically, we consider sparse deep neural network architecture with rectifier linear unit (ReLU) activation function and minimize the cross-entropy loss in the multiclass classification setup. This neural network architecture allows us to employ modern computational tools in the implementation. The convergence rates of the misclassification risk functions are also derived for both fully observed and discretely observed multidimensional functional data. We demonstrate the performance of mfDNN on simulated data and several benchmark datasets from different application domains.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is multi - class classification on multi - dimensional function data. Specifically, the paper focuses on how to achieve effective multi - class classification in multi - dimensional function data (such as 2D or 3D image data). Most of the existing classification methods are mainly applicable to one - dimensional function data, such as temperature data or speech recognition data, and there is less research on the classification of multi - dimensional function data. In addition, the existing methods can usually only handle binary classification problems and require fully observed data. This paper proposes a new multi - class functional deep neural network (mfDNN) classifier, aiming to overcome these limitations and be able to perform multi - class classification on partially observed or multi - dimensional function data. ### Main contributions: 1. **Proposing the mfDNN model**: The paper introduces a multi - class classifier based on a sparse deep neural network architecture, using the ReLU activation function and minimizing the cross - entropy loss in a multi - class classification setting. 2. **Theoretical analysis**: The paper derives the convergence rate of the misclassification risk function under fully observed and discretely observed multi - dimensional function data. 3. **Empirical analysis**: Through experiments on simulated data and multiple benchmark datasets, the performance of mfDNN is demonstrated and compared with existing classification methods. ### Specific problems solved: - **Multi - class classification of multi - dimensional function data**: Most of the existing methods are for one - dimensional function data, while the mfDNN proposed in this paper can handle multi - dimensional function data, such as 2D or 3D image data. - **Handling of partially observed data**: The paper not only considers fully observed data but also partially observed data, which is more common in practical applications. - **Multi - class classification**: The existing methods can usually only perform binary classification, while mfDNN can handle multi - class classification problems. ### Method overview: - **Data pre - processing**: Extract the function principal component scores (FPCs) as input features. - **Model construction**: Use a sparse deep ReLU network architecture and minimize the cross - entropy loss. - **Theoretical guarantee**: Derive the convergence rate of the misclassification risk function and provide a theoretical analysis of conditional probabilities. - **Empirical verification**: Verify the effectiveness of mfDNN through simulated data and real - world datasets. ### Conclusion: Through theoretical analysis and empirical research, the paper proves the effectiveness and superiority of mfDNN in multi - class classification tasks of multi - dimensional function data, especially when dealing with partially observed data and multi - class classification problems.

Multiclass classification for multidimensional functional data through deep neural networks

Deep Neural Network Classifier for Multi-dimensional Functional Data

Functional data analysis using deep neural networks

Neural Networks as Functional Classifiers

Optimal Linear Discriminant Analysis for High-Dimensional Functional Data

Robust Deep Neural Network Estimation for Multi-dimensional Functional Data

Nonlinear functional regression by functional deep neural network with kernel embedding

Multimodal functional deep learning for multiomics data

SFMD‐X: A New Functional Data Classifier Based on Shrinkage Functional Mahalanobis Distance

HCFNN: High-order Coverage Function Neural Network for Image Classification

Functional Sufficient Dimension Reduction for Functional Data Classification

Convolutional Neural Network with Sparse Strategies to Classify Dynamic Functional Connectivity.

Deep Learning With Functional Inputs

Deep Learning for Functional Data Analysis with Adaptive Basis Layers

Minimax optimal high-dimensional classification using deep neural networks

Fusion of effective dimension reduction and discriminative dictionary learning for high-dimensional classification

Estimation of the Mean Function of Functional Data via Deep Neural Networks

Exploiting Multi-Channels Deep Convolutional Neural Networks for Multivariate Time Series Classification

Multimodal hyper-connectivity of functional networks using functionally-weighted LASSO for MCI classification.

Dynamic Multivariate Functional Data Modeling Via Sparse Subspace Learning

High Dimensional Multispectral Image Fusion: Classification by Neural Network