Standard Tree Research Articles

Multiset-Trie Data Structure

This paper proposes a new data structure, multiset-trie, that is designed for storing and efficiently processing a set of multisets. Moreover, multiset-trie can operate on a set of sets without efficiency loss. The multiset-trie structure is a search tree with properties similar to those of a trie. It implements all standard search tree operations together with the multiset containment operations for searching sub-multisets and super-multisets. Suppose that we have a set of multisets S and a multiset X. The multiset containment operations retrieve multisets from S that are either sub-multisets or super-multisets of X. We present the mathematical analysis of a multiset-trie that gives the time complexity of the algorithms and the space complexity of the data structure. Further, the empirical analysis of the data structure is implemented in a series of experiments. The experiments illuminate the time complexity space of the multiset containment operations.

A Network Security Prediction Method Based on Attack Defense Tree

Modern networks are vulnerable to coordinated assaults. To perform threat analysis on individual and combined attack behaviors, it is essential to establish a security situation analysis model based on offensive and defensive decisions. The standard attack tree model used in this study ignores the impact of defenses, and the defensive tree model is not well scalable. Game theory is introduced into the attack tree model to describe specific network attack and defense event scenarios. First, analyze the logical relationship of different levels of attack behaviors in the network, integrate the offensive and defensive trees corresponding to different levels of attack events, to obtain a complete network offensive and defensive behavior tree, and then build a network offensive and defensive behavior tree model. Second, the basic offensive and defensive behavior tree extends from network offensive and defensive behavior, network detection equipment, and network defense measures. Propose an algorithm for the success rate of the attack target and calculate its attack probability. Based on these findings, we assess the potential for an attack and investigate the current state of network defenses. An offensive and defensive behavior tree model based on the BGP (Border Gateway protoc01) attack tree was developed in order to test the viability and efficacy of the network offensive and defensive behavior tree model. The network attack and defensive behavior tree model brought out in this study is capable of performing a wide range of calculations. The effect of defensive measures, as well as the ability to add and delete offensive and defensive behaviors at any node, has strong scalability and can provide a scientific basis for decision-making for network managers and operators.

Integrating Information Gain methods for Feature Selection in Distance Education Sentiment Analysis during Covid-19

The disparities in access to public assistances between rural and urban areas are analyzed in this study, corresponding to the intensity in which the constructs are related to expected quality, perceived quality and perceived value as influential factors in citizen satisfaction and loyalty. In Guayaquil, a survey is carried out at the household level, 428 valid questionnaires are obtained in the rural area of Tenguel and 521 Valid questionnaires in the urban area of Tarqui, applyingto the American Customer Satisfaction Index (ACSI). The research used a Structural Equation model (SEM) to evaluate the hypotheses raised, if they observe significant differences in citizen perception between inhabitants of urban and rural areas about the quality of community services, this as a determining cause in the level of citizen satisfaction and loyalty to choose their municipal authorities. The multigroup analysis allowed to identify inequalities in the observation of the quality of municipal or local communityassistances between rural and urban areas; the findings are considered to local public administrators for the design of public policy aimed at improving levels of citizen satisfaction and loyalty.Sentiment analysis is a way to automatically understand and process text data to figure out how someone feels about an opinion sentence. If there are too many reviews, it will take a lot of time and they will start to be biased. Sentiment classification tries to solve this problem by putting user reviews into groups based on whether they are positive, negative, or neutral. The dataset comes from Drone Emprit Academic. It is made up of tweets with the words "online learning method" in them, with as many as 4887 data crawled from them. Information Gain and adaboost on the C4.5 (FS+C4.5) method are used in the feature selection method. We use feature options to get rid of bias and improve accuracy. The results of the experiments will be compared to other algorithms like C4.5 and random forest. Based on the results, the accuracy of the two standard decision tree models (C4.5 and random forest) went up from 48.21% and 50.35% to 94.47 %. The value of how accurate it was went up by 44 percent. The FS+C4.5 model, on the other hand, has an RMSE of 0.204 and a correlation of 0.944. So, adding the feature selection technique to the sentiment analysis of bold learning education can make the C4.5 algorithm even more accurate.

Evaluating impacts of syntenic block detection strategies on rearrangement phylogeny using Mycobacterium tuberculosis isolates.

The phylogenetic signal of structural variation informs a more comprehensive understanding of evolution. As (near-)complete genome assembly becomes more commonplace, the next methodological challenge for inferring genome rearrangement trees is the identification of syntenic blocks of orthologous sequences. In this article, we studied 94 reference quality genomes of primarily Mycobacterium tuberculosis (Mtb) isolates as a benchmark to evaluate these methods. The clonal nature of Mtb evolution, the manageable genome sizes, along with substantial levels of structural variation make this an ideal benchmarking dataset. We tested several methods for detecting homology and obtaining syntenic blocks and two methods for inferring phylogenies from them, then compared the resulting trees to the standard method's tree, inferred from nucleotide substitutions. We found that, not only the choice of methods, but also their parameters can impact results, and that the tree inference method had less impact than the block determination method. Interestingly, a rearrangement tree based on blocks from the Cactus whole-genome aligner was fully compatible with the highly supported branches of the substitution-based tree, enabling the combination of the two into a high-resolution supertree. Overall, our results indicate that accurate trees can be inferred using genome rearrangements, but the choice of the methods for inferring homology requires care. Analysis scripts and code written for this study are available at https://gitlab.com/LPCDRP/rearrangement-homology.pub and https://gitlab.com/LPCDRP/syntement. Supplementary data are available at Bioinformatics online.

Robustan algoritam za učenje pravila za klasifikaciju - studija slučaja dijagnoze greške

Machine learning algorithms are used for building classifier models. The rule-based decision tree classifiers are popular ones. However, the performance of the decision tree classifier varies with hyperparameter tuning. The optimum hyperparameter values are obtained using either optimization algorithms or trial and error methods. The present study utilizes the MODLEM algorithm to overcome the drawbacks accounted for by decision tree algorithms. Eliminating hyperparameter tuning and producing results closer to standard decision tree algorithms makes MODLEM a robust classification algorithm. The robustness of the MODLEM algorithm is illustrated with the fault diagnosis case study. The case study is faults diagnosis of an automobile suspension system using vibration signals acquired at various fault conditions.

Heart Disease Classification for Early Diagnosis based on Adaptive Hoeffding Tree Algorithm in IoMT Data

Heart disease is a rapidly increasing disease that causes death worldwide. Therefore, scientists around the globe start studying this issue from a different perspective to assure early prediction of diagnosis to save patients' life from bad consequences that cause death. In this regard, Internet of Medical Things (IoMT) applications and algorithms should be utilized effectively to overcome this problem. Hoeffding Tree Algorithm (HTA) is a standard decision tree algorithm to handle large sizes of data sets. In this paper, an Adaptive Hoeffding Tree (AHT) algorithm is suggested to carry out classifications of data sets for early diagnosis of heart disease-related factors, and the obtained results by this algorithm are compared with other suggested Machine Learning (ML) algorithms in the literature. Therefore, a total of 3000 records of data sets are used in the classification, 33% of the data are utilized for female patient information, and the rest of the data are utilized for male patient information. In the original data set, each patient record includes 76 attributes, however only the most important 16 patient attributes are used for the classification. Data are retrieved from the University of California Irvine (UCI) Machine Learning Repository, which is collected from the Hungarian Institute of Cardiology, University Hospital at Zurich, University Hospital at Basel, and V.A. Medical Center. The obtained results from this study and the provided comparative results show the effectiveness of the AHT algorithm over other ML algorithms. Compared to other ML algorithms, AHT outperforms other algorithms with 95.67% accuracy for early estimation of diagnosis of heart disease.

S107 A Readmission Predictor Model to Identify High-Risk Patients With Crohn’s Disease

Background: Crohn’s disease is a chronic inflammatory autoimmune disease with unknown etiology that most commonly affects the younger population and females. Hospital readmission is frequent, burdensome, and costly. We investigated 30-day readmission rates, causes, risk factors, and interventions to reduce hospital readmission in patients who received medical treatment for Crohn’s disease. Methods: The Nationwide Readmission Database (HCUP) was queried from 2019-2022. Data on hospital readmissions of 1,048,576 adults who were readmitted within 30 days was collected. Our study first applied standard logistic regression and decision trees to obtain influential variables and derive practically meaningful decision rules. We then stratified the original data set and applied logistic regression to each data stratum. Finally, using area under the curve and odds ratio, we further explored the risk and accuracy of interacting variables in the logistic regression modeling. Results: A total of 1,048,576 patients were readmitted between 2019-2022. Of these, 15,438 (Mean age 56.4 ± 12.4, 57% women) patients were included after the propensity score matching. 7519 (48%) patients were Crohn’s positive. Multiple logistic regression of the independent variable showed a readmission probability of 8.3% in bottom quartile income (P < 0.01), 10.9% in HLD on medication (P < 0.01), 7.6% in COPD (P < 0.01), 6.8% in sepsis (P < 0.01), 4.8% in heart failure (P < 0.01), and 5.2% in insured patients (P < 0.01). The odds of readmission were increased in patients with a history of hypertension requiring medication (OR 1.201, P < 0.01), HLD requiring medication (OR 1.758, P < 0.01), COPD (OR 1.25, P < 0.01), sepsis (OR 1.10, P < 0.01), bottom quartile of income (OR 1.29, P < 0.01), and insured status (OR 1.03, P < 0.01). In addition, the female sex was associated with higher odds of readmission (OR 1.01, P < 0.01). Accuracy of bottom quartile income was also significant for AU ROC (0.664, P < 0.01) and insured status AU ROC (0.564, P < 0.01) as compared to the logistic regression. Conclusion(s): Our results suggest that patients with certain comorbid medical conditions, insurance status, and in bottom quartile income are at increased risk for readmission. Research is needed to determine if targeted interventions for high-risk patients bdecrease readmissions.

S108 Predictor Model to Identify High-Risk Groups for Readmission in Patients With Ulcerative Colitis

Background: Ulcerative colitis is characterized by recurring episodes of inflammation limited to the colon's mucosal layer. It commonly involves the rectum and may extend proximally and continuously to affect other parts of the colon. Risk factors for readmission over the first few weeks may differ from those that influence re-hospitalization at later time points. We investigated 30-day readmission rates, causes, risk factors, and interventions to reduce hospital readmission in patients who received medical treatment for ulcerative colitis. Methods: The Nationwide Readmission Database (HCUP) was queried from 2019-2022. Data on hospital readmissions of 1,048,576 adults who were readmitted within 30 days was collected. Our study first applied standard logistic regression and decision trees to obtain influential variables and derive practically meaningful decision rules. We then stratified the original data set and applied logistic regression to each data stratum. Finally, using the area under the curve and odds ratio, we further explored the risk and accuracy of interacting variables in the logistic regression modeling. Results: A total of 1,048,576 patients were readmitted between 2019-2022. Of these, 13,358 (Mean age 52.4 ± 10.4, 52% women) patients were included after the propensity score matching. 6215 (46%) patients were ulcerative colitis positive. Multiple logistic regression of the independent variable showed a readmission probability of 6.4% in the bottom quartile income group (P < 0.01), 5.3% in HLD on medication (P < 0.01), 1.3% in sepsis (P < 0.01), 4.7% in heart failure (P < 0.01), and 8.4% in insured patients (P < 0.01). The odds of readmission were increased in patients with a history of HLD requiring medication (OR 1.203, P < 0.01), heart failure (OR 1.303, P < 0.01), sepsis (OR 2.021, P < 0.01), bottom quartile of income (OR 1.45, P < 0.01), and insured status (OR 1.33, P < 0.01). In addition, the female sex was associated with higher odds of readmission (OR 1.21, P < 0.01). Accuracy of median income was also significant for AU ROC (0.632, P < 0.01) and insured status AU ROC (0.511, P < 0.01) as compared to the logistic regression. Conclusion(s): Our results suggest that patients with comorbid medical conditions, insured status, and bottom quartile of income at increased risk for readmission. Research is needed to determine if targeted interventions for high-risk patients decrease readmissions.

Gradient Tree Boosting for Hierarchical Data

Gradient tree boosting is a powerful machine learning technique that has shown good performance in predicting a variety of outcomes. However, when applied to hierarchical (e.g., longitudinal or clustered) data, the predictive performance of gradient tree boosting may be harmed by ignoring the hierarchical structure, and may be improved by accounting for it. Tree-based methods such as regression trees and random forests have already been extended to hierarchical data settings by combining them with the linear mixed effects model (MEM). In the present article, we add to this literature by proposing two algorithms to estimate a combination of the MEM and gradient tree boosting. We report on two simulation studies that (i) investigate the predictive performance of the two MEM boosting algorithms and (ii) compare them to standard gradient tree boosting, standard random forest, and other existing methods for hierarchical data (MEM, MEM random forests, model-based boosting, Bayesian additive regression trees [BART]). We found substantial improvements in the predictive performance of our MEM boosting algorithms over standard boosting when the random effects were non-negligible. MEM boosting as well as BART showed a predictive performance similar to the correctly specified MEM (i.e., the benchmark model), and overall outperformed the model-based boosting and random forest approaches.

First Evidence of a Combination of Terpinen-4-ol and α-Terpineol as a Promising Tool against ESKAPE Pathogens.

Antimicrobial resistance is a major public health issue raising growing concern in the face of dwindling response options. It is therefore urgent to find new anti-infective molecules enabling us to fight effectively against ever more numerous bacterial infections caused by ever more antibiotic-resistant bacteria. In this quest for new antibacterials, essential oils (or compounds extracted from essential oils) appear to be a promising therapeutic option. In the present work, we investigate the potential antibacterial synergy between a combination of terpinen-4-ol and α-terpineol (10:1) compared to standard tea tree oil. The minimum inhibitory concentration (MIC) and the minimum bactericidal concentration (MBC) were determined. Then, time kill assays, in vitro cytotoxicity and bactericidal activity on latent bacteria (persisters) were investigated. Finally, an in silico study of the pharmacokinetic parameters of α-terpineol was also performed. Altogether, our data demonstrate that the combination of terpinen-4-ol and α-terpineol might be a precious weapon to address ESKAPE pathogens.

Standard Tree and Shrub Plants in the Landscape Design of the City of Kyiv: Current Condition and Perspectives of Use

The purpose of the research is to summarize the information and evaluate the prerequisites for optimizing the taxonomic composition of plants for the general use plantations of the city of Kyiv. The work aims to provide specific proposals for the introduction of standard tree and shrub plants in the city’s public plantings, focusing on the motivation of plant selection based on ecological, biological and decorative features. Research methodology. In the process of conducting research, several special and generally scientific methods were used, including field survey, monitoring, description, analysis, synthesis, abstraction, generalization, comparison, forecasting and classification. The scientific novelty of the research consists in conducting an in-depth analysis of the modern taxonomic composition of standard tree and shrub plants, summarizing the condition and prerequisites for their use in the landscape design of Kyiv. Based on the results of the work, practical recommendations have been provided in the form of a specific list of standard tree and shrub plants, which are advisable to use in public plantations of the city, which confirms the practical significance of the conducted research. Conclusions. It has been established that modern landscape design trends, in particular, the use of standard tree and shrub plants, contribute to the improvement and modernization of public plantings in the city of Kyiv. The proposed list of these species’ plants aims to significantly enrich the taxonomic composition of green spaces and improve the urbanized space of the city. The motivation for the choice becomes quite clear given the decorative indicators of these plants, their biological and ecological stability, adaptability, dimensions (stem height, shape and size of the crown), safety and practicality of use in urban conditions. An essential enrichment of the standard tree and shrub plants’ taxonomic composition over the past decades has become possible thanks to active selection work abroad. Predictions in the field of landscape design and decorative horticulture testify to the planned changes in the system of green spaces in the city of Kyiv, including public spaces, under the influence of European experience.

Response of the fine root morphological and chemical traits of Tamarix chinensis to water and salt changes in coastal wetlands of the Yellow River Delta.

To explore the adaptation of the fine root morphology and chemical characteristics of Tamarix chinensis to water–salt heterogeneity in the groundwater–soil system of a coastal wetland zone, T. chinensis forests at different groundwater levels (high: GW1 0.54 m and GW2 0.83 m; medium: GW3 1.18 m; low: GW4 1.62 m and GW5 2.04 m) in the coastal wetland of the Yellow River Delta were researched, and the fine roots of T. chinensis standard trees were excavated. The fine roots were classified by the Pregitzer method, and the morphology, nutrients, and nonstructural carbohydrate characteristics of each order were determined. The results showed that the groundwater level had a significant indigenous effect on the soil water and salt conditions and affected the fine roots of T. chinensis. At high groundwater levels, the specific root length and specific surface area of fine roots were small, the root tissue density was high, the fine root growth rate was slow, the nutrient use efficiency was higher than at low groundwater levels, and the absorption of water increased with increasing specific surface area. With decreasing groundwater level, the N content and C/N ratio of fine roots first decreased and then increased, and the soluble sugar, starch, and nonstructural carbohydrate content of fine roots first increased and then decreased. At high and low groundwater levels, the metabolism of fine roots of T. chinensis was enhanced, and their adaptability to high salt content and low water content soil environments improved. The first- and second-order fine roots of T. chinensis were mainly responsible for water and nutrient absorption, while the higher-order (from the third to fifth orders) fine roots were primarily responsible for the transportation and storage of carbohydrates. The fine root morphology, nutrients, nonstructural carbohydrate characteristics, and other aspects of the water and salt environment heterogeneity cooperated in a synergistic response and trade-off adjustment.

PbHCT4 regulates growth through affecting chlorogenic acid (CGA) content in pear

• The chlorogenic acid contents between dwarf and standard pear trees were different significantly. • Low concentration of chlorogenic acid promoted the growth of pear, while high concentration of chlorogenic acid inhibited the growth. • PbHCT4 gene changed plant height by regulating chlorogenic acid content and auxin transport. Due to its many advantages, dwarfing and dense planting is an important mode of modern fruit tree cultivation. Phenols have been reported to play various roles in the growth regulation of plants. Here, to verify the correlation between chlorogenic acid (CGA) and the growth of pear trees, the CGA content of standard and dwarf pear trees as well as the effect of exogenous CGA upon their growth, were investigated and analyzed. Our results showed that the CGA content of two dwarf-type varieties, ‘Zhongai 1′ (Z1) and ‘PY-9′ was significantly higher than that of two standard-type Pyrus betulaefolia (P1 and P2). Whereas a low concentration of CGA accelerated pear tree growth, and a high concentration of CGA could inhibit it. Then, through an expression pattern analysis of four CGA synthesis-related genes, PbHCT1–4 , we found that the dwarf phenotype might be related to the expression levels of PbHCTs , especially to those of PbHCT2–4 in shoots. Of these, PbHCT4 was selected as a candidate gene and transformed into tobacco. When compared with control tobacco plants, the transgenic plants attained taller heights and had a greater CGA content, as well as increased expression levels of the auxin efflux carrier gene, NbPIN. Accordingly, we speculate the PbHCT4 gene promotes the synthesis of CGA, which influences plant growth by mediating auxin's transport in pear trees. These findings are helpful for advancing analysis of dwarf mechanism, which is critical for the breeding of new dwarf pear varieties.

Optimizing quality wood production in chestnut (Castanea sativa Mill.) coppices

In the last decades, releasing standards have become a very common and in some cases even mandatory management options in coppice systems in Europe. As for the chestnut tree, a light demanding and fast-growing species, however, there is a lively debate and a lack of scientific evidences about the pros and cons of releasing standards, especially in stands devoted to quality wood production. In this paper we used nine chestnut coppice stands growing on similar site conditions - but differing in stool density and number of standard trees released - to analyze the effects of the stand structure (i.e., stool and standard density) on the growing performance in terms of diameter increment of the dominant shoots and overall basal area production. Simple coppicing (i.e., no standards release) confirmed to be the most suitable system to enable an initial full growing performance of a light-demanding species such as the chestnut. We thus recommend avoiding the release of standards in quality-wood chestnut coppices in order to allow the new shoots generation to develop undisturbed until the stage of the first thinning. The release of standards in chestnut coppices may however be appropriate in particular cases, such as in stands whose main aim is to protect against shallow landslides along steep slopes.

Direct effects of tephra fallout from the Puyehue–Cordón Caulle Volcanic Complex on Nothofagus pumilio ring widths in northern Patagonia

We evaluated the radial growth response of adult Nothofagus pumilio (Poepp. et Endl) Krasser trees affected by tephra deposition following historical volcanic eruptions of the Puyehue–Cordón Caulle Volcanic Complex (PCCVC) in northern Patagonia. Standard tree–ring width chronologies were developed for trees from two sites that were affected by up to 55 cm of tephra during the 2011 eruption, which allowed us to detect the general tree–growth response to eruptions VEI ≥ 3 and VEI ≤ 2. The tree growth trend satisfactorily followed the mean temperature record (r = 0.42); however, the analysis of studentized residuals identified outliers (≥ ± 2 SD) directly related to the volcanic eruptions of the years 1921–1922 and 2011 and the respective post–eruption years, while for the 1960 eruption and following year, they largely exceeded the mean value of the residuals. The large amount of tephra deposited during the 1921–22 and 2011 eruptions caused physical damage to the tree canopy leading to the appearance of white rings and to locally absent rings. The rate of change in radial growth of trees during these eruptions presented significant declines in relation to the growth of five years before the eruption and to the following year. The low amount of tephra deposited during the 1960 eruption did not cause damage to the stands and trees increased their radial growth. In general, trees that had reduced radial growth experienced a remarkable recovery starting in the second or third post–eruption year. The amount of tephra deposited and the time of year of the volcanic eruptions had an important influence on tree rings. Some ecophysiological causes that could explain the growth responses of N. pumilio to tephra fall are discussed herein. Our study may provide useful insights to clarify the uncertain characteristics of some eruptions in the past or to detect the occurrence of large, undocumented volcanic eruptions throughout the Andes.

Identification of Old-Growth Mediterranean Forests Using Airborne Laser Scanning and Geostatistical Analysis

The protection and conservation of old-growth forests (OGFs) are becoming a global concern due to their irreplaceability and high biodiversity. Nonetheless, there has been little research into the identification and characterization of OGFs of the oldest tree species in Mediterranean areas. We used forest inventory data, low-density airborne laser scanning (ALS) metrics, and geostatistical analysis to estimate old-growth indices (OGIs) as indicators of old-growth forest conditions. We selected a pilot area in European black pine (Pinus nigra subsp. salzmannii) ecosystems where the oldest known living trees in the Iberian Peninsula are found. A total of 756 inventory plots were established to characterize standard live tree and stand attributes. We estimated several structural attributes that discriminate old growth from younger age classes and calculated different types of OGI for each plot. The best OGI was based on mean tree diameter, standard deviation of tree diameter, and stand density of large trees (diameter &gt; 50 cm). This index is useful for assessing old-growthness at different successional stages (young and OGFs) in Mediterranean black pine forests. Our results confirm that the estimation of OGIs based on a combination of forest inventory data, geostatistical analysis, and ALS is useful for identifying OGFs.

Pneumonia Detection in Chest X-Ray Images Using Enhanced Restricted Boltzmann Machine.

The process of pneumonia detection has been the focus of researchers as it has proved itself to be one of the most dangerous and life-threatening disorders. In recent years, many machine learning and deep learning algorithms have been applied in an attempt to automate this process but none of them has been successful significantly to achieve the highest possible accuracy. In a similar attempt, we propose an enhanced approach of a deep learning model called restricted Boltzmann machine (RBM) which is named enhanced RBM (ERBM). One of the major drawbacks associated with the standard format of RBM is its random weight initialization which leads to improper feature learning of the model during the training phase, resulting in poor performance of the machine. This problem has been tried to eliminate in this work by finding the differences between the means of a specific feature vector and the means of all features given as inputs to the machine. By performing this process, the reconstruction of the actual features is increased which ultimately reduces the error generated during the training phase of the model. The developed model has been applied to three different datasets of pneumonia diseases and the results have been compared with other state of the art techniques using different performance evaluation parameters. The proposed model gave highest accuracy of 98.56% followed by standard RBM, SVM, KNN, and decision tree which gave accuracies of 97.53%, 92.62%, 91.64%, and 88.77%, respectively, for dataset named dataset 2. Similarly, for the dataset 1, the highest accuracy of 96.66 has been observed for the eRBM followed by srRBM, KNN, decision tree, and SVM which gave accuracies of 90.22%, 89.34%, 87.65%, and 86.55%, respectively. In the same way, the accuracies observed for the dataset 3 by eRBM, standard RBM, KNN, decision tree, and SVM are 92.45%, 90.98%, 87.54%, 85.49%, and 84.54%, respectively. Similar observations can also be seen for other performance parameters showing the efficiency of the proposed model. As revealed in the results obtained, a significant improvement has been observed in the working of the RBM by introducing a new method of weight initialization during the training phase. The results show that the improved model outperforms other models in terms of different performance evaluation parameters, namely, accuracy, sensitivity, specificity, F1-score, and ROC curve.

Evidence for evolutionary relationship between archaeplastidal and cyanobacterial phytochromes based on their chromophore pockets

Phytochromes are photoreceptor proteins with a bilin chromophore that undergo photoconversion between two spectrally different forms, Pr and Pfr. In plants, phytochromes play a central role in growth and differentiation during the entire life cycle. Phytochromes of plants and other groups of archaeplastida have a common evolutionary origin in prokaryotes, but the exact prokaryotic origin is as yet uncertain. Two possibilities are presently discussed: either, archaeplastidal phytochromes arose from the last eukaryotic common ancestor (LECA) or they arose from the cyanobacterial endosymbiont that gave rise to plastids. We first constructed standard phylogenetic trees based on N-terminal protein sequences of the chromophore module. As usual, variation of algorithms and parameters led to different trees. A relationship between cyanobacteria and archaeplastida was observed in 7 out of 36 trees. The lack of consistency between results obtained from variation of parameters of tree constructions reflects the uncertainty of archaeplastidal origin. To gain more information about a possible cyanobacterial and archaeplastidal relationship, we performed phylogenetic studies based on the amino acids that line the chromophore pockets. These amino acids are highly conserved and could provide more accurate information about long evolutionary time scales, but the reduction of traits could also lead to insignificant results. From 30 selected chromophore-binding amino acids, 6 were invariant. The subsequent studies were thus based on the information dependent on 24 or fewer amino acid positions. Again, multiple trees were constructed to get information about the robustness of relationships. The very low number of information-containing traits resulted in low bootstrap values and many indistinguishable leaves. However, the major groups fungi, bacteria, cyanobacteria, and plants remained united. Without exception, cyanobacteria and archaeplastida were always closely linked. In this respect, the results were more robust than those of the classic approach, based on long contiguous sequences. We therefore consider cyanobacteria as the most likely origin of archaeplastidal phytochromes.

Beyond robustness: Resilience verification of tree-based classifiers

In this paper we criticize the robustness measure traditionally employed to assess the performance of machine learning models deployed in adversarial settings. To mitigate the limitations of robustness, we introduce a new measure called resilience and we focus on its verification. In particular, we discuss how resilience can be verified by combining a traditional robustness verification technique with a data-independent stability analysis, which identifies a subset of the feature space where the model does not change its predictions despite adversarial manipulations. We then introduce a formally sound data-independent stability analysis for decision trees and decision tree ensembles, which we experimentally assess on public datasets and we leverage for resilience verification. Our results show that resilience verification is useful and feasible in practice, yielding a more reliable security assessment of both standard and robust decision tree models.

Rare-event Simulation for Neural Network and Random Forest Predictors

We study rare-event simulation for a class of problems where the target hitting sets of interest are defined via modern machine learning tools such as neural networks and random forests. This problem is motivated from fast emerging studies on the safety evaluation of intelligent systems, robustness quantification of learning models, and other potential applications to large-scale simulation in which machine learning tools can be used to approximate complex rare-event set boundaries. We investigate an importance sampling scheme that integrates the dominating point machinery in large deviations and sequential mixed integer programming to locate the underlying dominating points. Our approach works for a range of neural network architectures including fully connected layers, rectified linear units, normalization, pooling and convolutional layers, and random forests built from standard decision trees. We provide efficiency guarantees and numerical demonstration of our approach using a classification model in the UCI Machine Learning Repository.