Topological data analysis (TDA) is a branch of computational mathematics, bridging algebraic topology and data science, that provides compact, noise-robust representations of complex structures. Deep neural networks (DNNs) learn millions of parameters associated with a series of transformations defined by the model architecture resulting in high-dimensional, difficult to interpret internal representations of input data. As DNNs become more ubiquitous across multiple sectors of our society, there is increasing recognition that mathematical methods are needed to aid analysts, researchers, and practitioners in understanding and interpreting how these models' internal representations relate to the final classification. In this paper we apply cutting edge techniques from TDA with the goal of gaining insight towards interpretability of convolutional neural networks used for image classification. We use two common TDA approaches to explore several methods for modeling hidden layer activations as high-dimensional point clouds, and provide experimental evidence that these point clouds capture valuable structural information about the model's process. First, we demonstrate that a distance metric based on persistent homology can be used to quantify meaningful differences between layers and discuss these distances in the broader context of existing representational similarity metrics for neural network interpretability. Second, we show that a mapper graph can provide semantic insight as to how these models organize hierarchical class knowledge at each layer. These observations demonstrate that TDA is a useful tool to help deep learning practitioners unlock the hidden structures of their models.
more »
« less
A Primer on Topological Data Analysis to Support Image Analysis Tasks in Environmental Science
Topological data analysis (TDA) is a tool from data science and mathematics that is beginning to make waves in environmental science. In this work, we seek to provide an intuitive and understandable introduction to a tool from TDA that is particularly useful for the analysis of imagery, namely, persistent homology. We briefly discuss the theoretical background but focus primarily on understanding the output of this tool and discussing what information it can glean. To this end, we frame our discussion around a guiding example of classifying satellite images from the sugar, fish, flower, and gravel dataset produced for the study of mesoscale organization of clouds by Rasp et al. We demonstrate how persistent homology and its vectorization, persistence landscapes, can be used in a workflow with a simple machine learning algorithm to obtain good results, and we explore in detail how we can explain this behavior in terms of image-level features. One of the core strengths of persistent homology is how interpretable it can be, so throughout this paper we discuss not just the patterns we find but why those results are to be expected given what we know about the theory of persistent homology. Our goal is that readers of this paper will leave with a better understanding of TDA and persistent homology, will be able to identify problems and datasets of their own for which persistent homology could be helpful, and will gain an understanding of the results they obtain from applying the included GitHub example code.
Information such as the geometric structure and texture of image data can greatly support the inference of the physical state of an observed Earth system, for example, in remote sensing to determine whether wildfires are active or to identify local climate zones. Persistent homology is a branch of topological data analysis that allows one to extract such information in an interpretable way—unlike black-box methods like deep neural networks. The purpose of this paper is to explain in an intuitive manner what persistent homology is and how researchers in environmental science can use it to create interpretable models. We demonstrate the approach to identify certain cloud patterns from satellite imagery and find that the resulting model is indeed interpretable.
- Award ID(s):
- 1934668
- NSF-PAR ID:
- 10398807
- Publisher / Repository:
- American Meteorological Society
- Date Published:
- Journal Name:
- Artificial Intelligence for the Earth Systems
- Volume:
- 2
- Issue:
- 1
- ISSN:
- 2769-7525
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)Abstract This paper introduces the use of topological data analysis (TDA) as an unsupervised machine learning tool to uncover classification criteria in complex inorganic crystal chemistries. Using the apatite chemistry as a template, we track through the use of persistent homology the topological connectivity of input crystal chemistry descriptors on defining similarity between different stoichiometries of apatites. It is shown that TDA automatically identifies a hierarchical classification scheme within apatites based on the commonality of the number of discrete coordination polyhedra that constitute the structural building units common among the compounds. This information is presented in the form of a visualization scheme of a barcode of homology classifications, where the persistence of similarity between compounds is tracked. Unlike traditional perspectives of structure maps, this new “Materials Barcode” schema serves as an automated exploratory machine learning tool that can uncover structural associations from crystal chemistry databases, as well as to achieve a more nuanced insight into what defines similarity among homologous compounds.more » « less
-
An emerging method for data analysis is called Topological Data Analysis (TDA). TDA is based in the mathematical field of topology and examines the properties of spaces under continuous deformation. One of the key tools used for TDA is called persistent homology which considers the connectivity of points in a d-dimensional point cloud at different spatial resolutions to identify topological properties (holes, loops, and voids) in the space. Persistent homology then classifies the topological features by their persistence through the range of spatial connectivity. Unfortunately the memory and run-time complexity of computing persistent homology is exponential and current tools can only process a few thousand points in R3. Fortunately, the use of data reduction techniques enables persistent homology to be applied to much larger point clouds. Techniques to reduce the data range from random sampling of points to clustering the data and using the cluster centroids as the reduced data. While several data reduction approaches appear to preserve the large topological features present in the original point cloud, no systematic study comparing the efficacy of different data clustering techniques in preserving the persistent homology results has been performed. This paper explores the question of topology preserving data reductions and describes formally when and how topological features can be mischaracterized or lost by data reduction techniques. The paper also performs an experimental assessment of data reduction techniques and resilient effects on the persistent homology. In particular, data reduction by random selection is compared to cluster centroids extracted from different data clustering algorithms.more » « less
-
more » « less
Abstract Background Interpretation of high-throughput gene expression data continues to require mathematical tools in data analysis that recognizes the shape of the data in high dimensions. Topological data analysis (TDA) has recently been successful in extracting robust features in several applications dealing with high dimensional constructs. In this work, we utilize some recent developments in TDA to curate gene expression data. Our work differs from the predecessors in two aspects: (1) Traditional TDA pipelines use topological signatures called barcodes to enhance feature vectors which are used for classification. In contrast, this work involves curating relevant features to obtain somewhat better representatives with the help of TDA. This representatives of the entire data facilitates better comprehension of the phenotype labels. (2) Most of the earlier works employ barcodes obtained using topological summaries as fingerprints for the data. Even though they are stable signatures, there exists no direct mapping between the data and said barcodes.
Results The topology relevant curated data that we obtain provides an improvement in shallow learning as well as deep learning based supervised classifications. We further show that the representative cycles we compute have an unsupervised inclination towards phenotype labels. This work thus shows that topological signatures are able to comprehend gene expression levels and classify cohorts accordingly.
Conclusions In this work, we engender representative persistent cycles to discern the gene expression data. These cycles allow us to directly procure genes entailed in similar processes.
-
more » « less
Abstract Accurate prediction of physical alterations in carbonate reservoirs under dissolution is critical for development of subsurface energy technologies. The impact of mineral dissolution on flow characteristics depends on the connectivity and tortuosity of the pore network. Persistent homology is a tool from algebraic topology that describes the size and connectivity of topological features. When applied to 3D X‐ray computed tomography (XCT) imagery of rock cores, it provides a novel metric of pore network heterogeneity. Prior works have demonstrated the efficacy of persistent homology in predicting flow properties in numerical simulations of flow through porous media. Its ability to combine size, spatial distribution, and connectivity information make it a promising tool for understanding reactive transport in complex pore networks, yet limited work has been done to apply persistence analysis to experimental studies on natural rocks. In this study, three limestone cores were imaged by XCT before and after acid‐driven dissolution flow‐through experiments. Each XCT scan was analyzed using persistent homology. In all three rocks, permeability increase was driven by the growth of large, connected pore bodies. The two most homogenous samples saw an increased effect nearer to the flow inlet, suggesting emerging preferential flow paths as the reaction front progresses. The most heterogeneous sample showed an increase in along‐core homogeneity during reaction. Variability of persistence showed moderate positive correlation with pore body size increase. Persistence heterogeneity analysis could be used to anticipate where greatest pore size evolution may occur in a reservoir targeted for subsurface development, improving confidence in project viability.
