Abstract:The fractal dimension provides a statistical index of object complexity by studying how the pattern changes with the measuring scale. Although useful in several classification tasks, the fractal dimension is under-explored in deep learning applications. In this work, we investigate the features that are learned by deep models and we study whether these deep networks are able to encode features as complex and high-level as the fractal dimensions. Specifically, we conduct a correlation analysis experiment to show that deep networks are not able to extract such a feature in none of their layers. We combine our analytical study with a human evaluation to investigate the differences between deep learning networks and models that operate on the fractal feature solely. Moreover, we show the effectiveness of fractal features in applications where the object structure is crucial for the classification task. We empirically show that training a shallow network on fractal features achieves performance comparable, even superior in specific cases, to that of deep networks trained on raw data while requiring less computational resources. Fractals improved the accuracy of the classification by 30% on average while requiring up to 84% less time to train. We couple our empirical study with a complexity analysis of the computational cost of extracting the proposed fractal features, and we study its limitation.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is to explore whether deep - learning models (especially convolutional neural networks, CNNs) can encode fractal geometric features. Specifically, the author studies this problem in the following aspects: 1. **Extraction of Fractal Dimensions and Correlation Analysis**: - The author proposes a method for extracting fractal features from digital images and uses classic correlation analysis methods (such as CCA and CKA) to evaluate the similarity between these fractal features and the hidden - layer representations of deep - learning models. - Formula representation: \[ \text{Fractal Dimension (FD)}=\frac{\ln(N)}{\ln(1 / r)} \] where \( N \) is the number of self - similar segments and \( r \) is the magnification factor. 2. **Experimental Verification**: - The author conducts experiments on multiple datasets, including steel defect detection, potato and tomato disease detection, etc., to verify whether deep - learning models can encode fractal features at different levels. - The experimental results show that the fractal feature encoding ability of deep - learning models at all levels is very limited, indicating that these models are difficult to capture the complexity of fractal geometry. 3. **Comparison between Shallow Models and Deep - Learning Models**: - The author further trains a shallow neural network (ZFrac + NN) based on fractal features and compares it with deep - learning models. - The results show that in some specific tasks, the shallow model based on fractal features can not only be comparable to deep - learning models in performance, but even outperform the latter in some cases, while requiring fewer computational resources. 4. **Human Evaluation**: - To further verify the effectiveness of fractal features, the author also conducts human evaluation experiments and finds that the model based on fractal features behaves more reasonably in some mispredictions, which indicates that fractal features have unique advantages in some tasks. In conclusion, this paper aims to reveal the limitations of deep - learning models in encoding fractal geometric features and explore the potential of fractal features in specific classification tasks.

On The Potential of The Fractal Geometry and The CNNs Ability to Encode it

Two-Stream Boundary-Aware Neural Network for Concrete Crack Segmentation and Quantification

Fractal Dimension Invariant Filtering and Its CNN-based Implementation

Encoding Spatial Distribution of Convolutional Features for Texture Representation.

Towards Better Analysis of Deep Convolutional Neural Networks

A Scale-Invariant Diagnostic Approach Towards Understanding Dynamics of Deep Neural Networks

Deep Fractal Residual Network for Fast and Accurate Single Image Super Resolution

Intrinsic Dimension, Persistent Homology and Generalization in Neural Networks

Uncovering Nested Data Parallelism and Data Reuse in DNN Computation with FractalTensor

CNN-G: Convolutional Neural Network Combined With Graph for Image Segmentation With Theoretical Analysis

Representing Topological Self-Similarity Using Fractal Feature Maps for Accurate Segmentation of Tubular Structures

Improving Fractal Pre-training

Refining Architectures of Deep Convolutional Neural Networks

An ensemble of learned features and reshaping of fractal geometry-based descriptors for classification of histological images

Recognition Of 3D Surface Fractal Dimension Based On Convolutional Neural Network

Deep Continuous Networks

A novel feature-scrambling approach reveals the capacity of convolutional neural networks to learn spatial relations

Fast hybrid fractal image compression using an image feature and neural network

Approximation analysis of CNNs from a feature extraction view

Genetic CNN

How neurons exploit fractal geometry to optimize their network connectivity