skip to main content


The NSF Public Access Repository (NSF-PAR) system and access will be unavailable from 10:00 PM ET on Friday, December 8 until 2:00 AM ET on Saturday, December 9 due to maintenance. We apologize for the inconvenience.

This content will become publicly available on June 27, 2024

Title: Learning Fractals by Gradient Descent
Fractals are geometric shapes that can display complex and self-similar patterns found in nature (e.g., clouds and plants). Recent works in visual recognition have leveraged this property to create random fractal images for model pre-training. In this paper, we study the inverse problem --- given a target image (not necessarily a fractal), we aim to generate a fractal image that looks like it. We propose a novel approach that learns the parameters underlying a fractal image via gradient descent. We show that our approach can find fractal parameters of high visual quality and be compatible with different loss functions, opening up several potentials, e.g., learning fractals for downstream tasks, scientific understanding, etc.  more » « less
Award ID(s):
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Proceedings of the AAAI Conference on Artificial Intelligence
Page Range / eLocation ID:
2456 to 2464
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Forensic photography, also referred to as crime scene photography, is an activity that records the initial appearance of the crime scene and physical evidence in order to provide a permanent record for the court. Nowadays, we cannot imagine a crime scene investigation without photographic evidence. Crime or accident scene photographs can often be reanalyzed in cold cases or when the images need to be enlarged to show critical details. Fractals are rough or fragmented geometric shapes that can be subdivided into parts, each of which is a reduced copy of the whole. Fractal dimension (FD) is an important fractal geometry feature. There are many applications of fractals in various forensic fields, including image processing, image analysis, texture segmentation, shape classification, and identifying the image features such as roughness and smoothness of an image. Fractal analysis is applicable in forensic archeology and paleontology, as well. The damaged image can be reviewed, analyzed, and reconstructed by fractal nature analysis. 
    more » « less
  2. The deep neural networks used in modern computer vision systems require enormous image datasets to train them. These carefully-curated datasets typically have a million or more images, across a thousand or more distinct categories. The process of creating and curating such a dataset is a monumental undertaking, demanding extensive effort and labelling expense and necessitating careful navigation of technical and social issues such as label accuracy, copyright ownership, and content bias.What if we had a way to harness the power of large image datasets but with few or none of the major issues and concerns currently faced? This paper extends the recent work of Kataoka et al. [15], proposing an improved pre-training dataset based on dynamically-generated fractal images. Challenging issues with large-scale image datasets become points of elegance for fractal pre-training: perfect label accuracy at zero cost; no need to store/transmit large image archives; no privacy/demographic bias/concerns of inappropriate content, as no humans are pictured; limitless supply and diversity of images; and the images are free/open-source. Perhaps surprisingly, avoiding these difficulties imposes only a small penalty in performance. Leveraging a newly-proposed pre-training task—multi-instance prediction—our experiments demonstrate that fine-tuning a network pre-trained using fractals attains 92.7-98.1% of the accuracy of an ImageNet pre-trained network. Our code is publicly available. 1 
    more » « less
  3. Today in the age of advanced ceramic civilization, there are a variety of applications for modern ceramics materials with specific properties. Our up-to date research recognizes that ceramics have a fractal configuration nature on the basis of different phenomena. The key property of fractals is their scale-independence. The practical value is that the fractal objects’ interaction and energy is possible at any reasonable scale of magnitude, including the nanoscale and may be even below. This is a consequence of fractal scale independence. This brings us to the conclusion that properties of fractals are valid on any scale (macro, micro, or nano). We also analyzed these questions with experimental results obtained from a comet, here 67P, and also from ceramic grain and pore morphologies on the microstructure level. Fractality, as a scale-independent morphology, provides significant variety of opportunities, for example for energy storage. From the viewpoint of scaling, the relation between large and small in fractal analysis is very important. An ideal fractal can be magnified endlessly but natural morphologies cannot, what is the new light in materials sciences and space. 
    more » « less
  4. Abstract We develop a method for calculating the persistence landscapes of affine fractals using the parameters of the corresponding transformations. Given an iterated function system of affine transformations that satisfies a certain compatibility condition, we prove that there exists an affine transformation acting on the space of persistence landscapes, which intertwines the action of the iterated function system. This latter affine transformation is a strict contraction and its unique fixed point is the persistence landscape of the affine fractal. We present several examples of the theory as well as confirm the main results through simulations. 
    more » « less
  5. Amancio, Diego R. (Ed.)
    Using fractal analyses to study events allows us to capture the scale-independence of those events, that is, no matter at which level we study a phenomenon, we should get roughly the same results because events exhibit similar structure across scales. This is demonstrably true in mathematical fractals but is less assured in behavioral fractals. The current research directly tests the scale-independence hypothesis in the behavioral domain by exploring the fractal structure of aggression, a social phenomenon comprising events that span temporal scales from minutes of face-to-face arguments to centuries of international armed conflicts. Using publicly available data, we examined the temporal fractal structure of four scales of aggression: wars (very macrolevel, worldwide data), riots (macrolevel, worldwide data), violent crimes (microlevel, data gathered from cities and towns in the United States of America), and body movement during arguments (very microlevel, data gathered on American participants). Our results lend mixed support to the scale-independence hypothesis and provide insight into the self-organization of human interactions. 
    more » « less