Topology-oriented self-organizing maps: a survey

Spherical Tree-Structured SOM and Its Application to Hierarchical Clustering

When analyzing high-dimensional data with many elements, a data visualization that maps the data onto a low-dimensional space is often performed. By visualizing the data, humans can intuitively understand the structure of the data in the high-dimensional space. The self-organizing map (SOM) is one such data visualization method. We propose a spherical tree-structured SOM (S-TS-SOM), which speeds up the search for winner nodes and eliminates the unevenness of learning due to the position of the winner nodes by placing the nodes on a sphere and applying the tree search method. In this paper, we confirm that the S-TS-SOM can achieve the same results as a normal spherical SOM while reducing the learning time. In addition, we confirm the granularity of clustering on the tree structure of the S-TS-SOM.

Supporting Clinical COVID-19 Diagnosis with Routine Blood Tests Using Tree-Based Entropy Structured Self-Organizing Maps

Data classification is an automatic or semi-automatic process that, utilizing artificial intelligence algorithms, learns the variable and class relationships of a dataset for use a posteriori in situations where the class result is unknown. For many years, work on this topic has been aimed at increasing the hit rates of algorithms. However, when the problem is restricted to applications in healthcare, besides the concern with performance, it is also necessary to design algorithms whose results are understandable by the specialists responsible for making the decisions. Among the problems in the field of medicine, a current focus is related to COVID-19: AI algorithms may contribute to early diagnosis. Among the available COVID-19 data, the blood test is a typical procedure performed when the patient seeks the hospital, and its use in the diagnosis allows reducing the need for other diagnostic tests that can impact the detection time and add to costs. In this work, we propose using self-organizing map (SOM) to discover attributes in blood test examinations that are relevant for COVID-19 diagnosis. We applied SOM and an entropy calculation in the definition of a hierarchical, semi-supervised and explainable model named TESSOM (tree-based entropy-structured self-organizing maps), in which the main feature is enhancing the investigation of groups of cases with high levels of class overlap, as far as the diagnostic outcome is concerned. Framing the TESSOM algorithm in the context of explainable artificial intelligence (XAI) makes it possible to explain the results to an expert in a simplified way. It is demonstrated in the paper that the use of the TESSOM algorithm to identify attributes of blood tests can help with the identification of COVID-19 cases. It providing a performance increase in 1.489% in multiple scenarios when analyzing 2207 cases from three hospitals in the state of São Paulo, Brazil. This work is a starting point for researchers to identify relevant attributes of blood tests for COVID-19 and to support the diagnosis of other diseases.

Challenges for a Massive Implementation of Phenomics in Plant Breeding Programs

Due to climate change and expected food shortage in the coming decades, not only will it be necessary to develop cultivars with greater tolerance to environmental stress, but it is also imperative to reduce breeding cycle time. In addition to yield evaluation, plant breeders resort to many sensory assessments and some others of intermediate complexity. However, to develop cultivars better adapted to current/future constraints, it is necessary to incorporate a new set of traits, such as morphophysiological and physicochemical attributes, information relevant to the successful selection of genotypes or parents. Unfortunately, because of the large number of genotypes to be screened, measurements with conventional equipment are unfeasible, especially under field conditions. High-throughput plant phenotyping (HTPP) facilitates collecting a significant amount of data quickly; however, it is necessary to transform all this information (e.g., plant reflectance) into helpful descriptors to the breeder. To the extent that a holistic characterization of the plant (phenomics) is performed in challenging environments, it will be possible to select the best genotypes (forward phenomics) objectively but also understand why the said individual differs from the rest (reverse phenomics). Unfortunately, several elements had prevented phenomics from developing as desired. Consequently, a new set of prediction/validation methodologies, seasonal ambient information, and the fusion of data matrices (e.g., genotypic and phenotypic information) need to be incorporated into the modeling. In this sense, for the massive implementation of phenomics in plant breeding, it will be essential to count an interdisciplinary team that responds to the urgent need to release material with greater capacity to tolerate environmental stress. Therefore, breeding programs should (i) be more efficient (e.g., early discarding of unsuitable material), (ii) have shorter breeding cycles (fewer crosses to achieve the desired cultivar), and (iii) be more productive, increasing the probability of success at the end of the breeding process (percentage of cultivars released to the number of initial crosses).

Randomized Self-Organizing Map

We propose a variation of the self-organizing map algorithm by considering the random placement of neurons on a two-dimensional manifold, following a blue noise distribution from which various topologies can be derived. These topologies possess random (but controllable) discontinuities that allow for a more flexible self-organization, especially with high-dimensional data. The proposed algorithm is tested on one-, two- and three-dimensional tasks, as well as on the MNIST handwritten digits data set and validated using spectral analysis and topological data analysis tools. We also demonstrate the ability of the randomized self-organizing map to gracefully reorganize itself in case of neural lesion and/or neurogenesis.

Sectoral performance analysis of national greenhouse gas emission inventories by means of neural networks

Annual greenhouse gas emissions have increased more than threefold between 1950 and 2014, posing a major threat to the integrity of the entire earth system and subsequently to humankind. Consequently, roadmaps towards low-carbon pathways are urgently needed. Our study contributes to a more detailed understanding of the dynamics of country based emission patterns and uses them to discuss prospective low-carbon pathways for countries. As availability of databases on sectoral emissions substantially increased, we employ machine learning techniques to classify emission features and pathways. By doing so, 18 representative emission patterns are derived. Overall emissions from seven sectors and for 167 countries covering the time span from 1950 to 2014 have been used in the analyses. The following significant trends can be observed: a) increasing per capita emissions due to growing fossil fuel use in many parts of the world, b) a decline in per capita emissions in some countries, and c) a shift in the emission shares, i.e., a reduction of agricultural and land use contributions in certain regions. Using the emission patterns, their dynamics, and best performing countries as role models, we show the possibility for gaining a decent human development without significantly increasing per capita emissions.

Organizing Books and Authors by Multilayer SOM

This paper introduces a new framework for the organization of electronic books (e-books) and their corresponding authors using a multilayer self-organizing map (MLSOM). An author is modeled by a rich tree-structured representation, and an MLSOM-based system is used as an efficient solution to the organizational problem of structured data. The tree-structured representation formulates author features in a hierarchy of author biography, books, pages, and paragraphs. To efficiently tackle the tree-structured representation, we used an MLSOM algorithm that serves as a clustering technique to handle e-books and their corresponding authors. A book and author recommender system is then implemented using the proposed framework. The effectiveness of our approach was examined in a large-scale data set containing 3868 authors along with the 10500 e-books that they wrote. We also provided visualization results of MLSOM for revealing the relevance patterns hidden from presented author clusters. The experimental results corroborate that the proposed method outperforms other content-based models (e.g., rate adapting poisson, latent Dirichlet allocation, probabilistic latent semantic indexing, and so on) and offers a promising solution to book recommendation, author recommendation, and visualization.

Evaluating the utility of mid-infrared spectral subspaces for predicting soil properties

We propose four methods for finding local subspaces in large spectral libraries. The proposed four methods include (a) cosine angle spectral matching; (b) hit quality index spectral matching; (c) self-organizing maps and (d) archetypal analysis methods. Then evaluate prediction accuracies for global and subspaces calibration models. These methods were tested on a mid-infrared spectral library containing 1907 soil samples collected from 19 different countries under the Africa Soil Information Service project. Calibration models for pH, Mehlich-3 Ca, Mehlich-3 Al, total carbon and clay soil properties were developed for the whole library and for the subspace. Root mean square error of prediction was used to evaluate predictive performance of subspace and global models. The root mean square error of prediction was computed using a one-third-holdout validation set. Effect of pretreating spectra with different methods was tested for 1st and 2nd derivative Savitzky-Golay algorithm, multiplicative scatter correction, standard normal variate and standard normal variate followed by detrending methods. In summary, the results show that global models outperformed the subspace models. We, therefore, conclude that global models are more accurate than the local models except in few cases. For instance, sand and clay root mean square error values from local models from archetypal analysis method were 50% poorer than the global models except for subspace models obtained using multiplicative scatter corrected spectra with which were 12% better. However, the subspace approach provides novel methods for discovering data pattern that may exist in large spectral libraries.