Making Classification Decisions Research Articles

Comparing human text classification performance and explainability with large language and machine learning models using eye-tracking

To understand the alignment between reasonings of humans and artificial intelligence (AI) models, this empirical study compared the human text classification performance and explainability with a traditional machine learning (ML) model and large language model (LLM). A domain-specific noisy textual dataset of 204 injury narratives had to be classified into 6 cause-of-injury codes. The narratives varied in terms of complexity and ease of categorization based on the distinctive nature of cause-of-injury code. The user study involved 51 participants whose eye-tracking data was recorded while they performed the text classification task. While the ML model was trained on 120,000 pre-labelled injury narratives, LLM and humans did not receive any specialized training. The explainability of different approaches was compared based on the top words they used for making classification decision. These words were identified using eye-tracking for humans, explainable AI approach LIME for ML model, and prompts for LLM. The classification performance of ML model was observed to be relatively better than zero-shot LLM and non-expert humans, overall, and particularly for narratives with high complexity and difficult categorization. The top-3 predictive words used by ML and LLM for classification agreed with humans to a greater extent as compared to later predictive words.

Application of convolutional neural network for differentiating ovarian thecoma-fibroma and solid ovarian cancer based on MRI.

Ovarian thecoma-fibroma and solid ovarian cancer have similar clinical and imaging features, and it is difficult for radiologists to differentiate them. Since the treatment and prognosis of them are different, accurate characterization is crucial. To non-invasively differentiate ovarian thecoma-fibroma and solid ovarian cancer by convolutional neural network based on magnetic resonance imaging (MRI), and to provide the interpretability of the model. A total of 156 tumors, including 86 ovarian thecoma-fibroma and 70 solid ovarian cancer, were split into the training set, the validation set, and the test set according to the ratio of 8:1:1 by stratified random sampling. In this study, we used four different networks, two different weight modes, two different optimizers, and four different sizes of regions of interest (ROI) to test the model performance. This process was repeated 10 times to calculate the average performance of the test set. The gradient weighted class activation mapping (Grad-CAM) was used to explain how the model makes classification decisions by visual location map. ResNet18, which had pre-trained weight, using Adam and one multiple ROI circumscribed rectangle, achieved best performance. The average accuracy, precision, recall, and AUC were 0.852, 0.828, 0.848, and 0.919 (P < 0.01), respectively. Grad-CAM showed areas associated with classification appeared on the edge or interior of ovarian thecoma-fibroma and the interior of solid ovarian cancer. This study shows that convolution neural network based on MRI can be helpful for radiologists in differentiating ovarian thecoma-fibroma and solid ovarian cancer.

Post-training flexibility in category learning.

Exemplar models of categorization, which assume that people make classification decisions based on item information stored in memory, typically assume that all of the exemplars are available and inform decision-making. However, in this study, we hypothesized that people may selectively emphasize subsets of exemplars, giving rise to individual differences in categorization. To verify this hypothesis, we adopted the partial-XOR category structure in Conaway and Kurtz (Psychonomic Bulletin & Review, 24, 1312-1323 2017), which has been evident to be able to induce two major response patterns in the transfer phase: the Proximity and XOR patterns. "Experiment 1" confirmed that these two patterns could be generated if participants were trained with only the exemplars of one category or the other. In "Experiment 2", participants were asked to not only learn the category labels of all exemplars but also memorize the exemplars of only Category A (Condition A), only Category B (Condition B), or two categories (Condition AB) for a recognition test after the training phase of the categorization task. As expected, in the transfer phase, the participants tended to perform the XOR and Proximity patterns, when the exemplars of Category A and Category B were respectively targeted for the recognition test. The parameters of the SDGCM estimated by Bayesian inference for modeling the data of "Experiment 2" showed that the exemplar accessibility of Category A was larger than that of Category B for performing the XOR pattern and vice versa for performing the proximity pattern, hence verifying our hypothesis.

A semi-supervised concept-cognitive computing system for dynamic classification decision making with limited feedback information

In dynamic environments, making classification decisions based on classical intelligent decision support systems is a challenge, as the classification performance of decision-making and the time-cost of learning need to be considered simultaneously. Moreover, many tasks of classification decisions lack label information because annotating data is time-consuming, labor-intensive and expensive process. This means that some standard intelligent decision support systems will perform inferior performance if they cannot dynamically make full use of the information behind abundant unlabeled data. Therefore, by incorporating knowledge representation and dynamic updating mechanisms into concept learning processes, we introduce a novel dynamic concept learning approach, namely semi-supervised concept-cognitive computing system (s2C3S), for making classification decisions by jointly utilizing some labeled data and abundant unlabeled data under dynamic environments. A theoretical analysis has shown that the proposed s2C3S can achieve significantly lower computational costs and higher classification accuracies than the existing incremental K Nearest Neighbor method (IKNN) and concept-cognitive computing system (C3S). The experimental results on various datasets further demonstrated that our system is effective for dynamic classification decision-making with limited labeled data under dynamic learning processes. Additionally, s2C3S can also be applied to computer-assisted intelligent diagnosis from the given medical images (such as chest X-ray images) dynamically and accurately.

Comparative Study of K-NN, Naive Bayes and SVM for Face Expression Classification Techniques

Abstract Classification is a data mining technique used to predict group membership for data instances within a given dataset. Classification is considered as an example of supervised learning as training data associated with class labels is given as input. This comparative study focuses on the application of K-Nearest Neighbors (K-NN), Naive Bayes, and Support Vector Machines (SVM) classification techniques in the context of face expression recognition. The study aims to evaluate and compare the performance of these techniques in accurately classifying facial expressions, their advantages and disadvantages. Facial expression recognition is an important area in computer vision and emotion analysis. It involves identifying and classifying different facial expressions such as happiness, sadness, anger, fear, surprise and disgust. Accurate classification of facial expressions is crucial for various applications, including human-computer interaction, affective computing and psychological research. K-NN is a non-parametric algorithm that classifies data based on the majority class of its K-nearest neighbors. Naive Bayes is a probabilistic algorithm that makes classification decisions based on the Bayes theorem and assumes independence between features. SVM is a machine learning algorithm that constructs a hyperplane to separate different classes of data. Dataset such as Affectnet is used to conduct the study including labeled facial expression images. Three algorithms are tested on this dataset to evaluate their performance in correctly classifying facial expressions. Performance metrics such as accuracy, precision, recall, and F1-score are used to assess the classification performance of each technique. Software used for conducting the experiments is Python.

Collaborative Decision-Reinforced Self-Supervision for Attributed Graph Clustering.

Attributed graph clustering aims to partition nodes of a graph structure into different groups. Recent works usually use variational graph autoencoder (VGAE) to make the node representations obey a specific distribution. Although they have shown promising results, how to introduce supervised information to guide the representation learning of graph nodes and improve clustering performance is still an open problem. In this article, we propose a Collaborative Decision-Reinforced Self-Supervision (CDRS) method to solve the problem, in which a pseudo node classification task collaborates with the clustering task to enhance the representation learning of graph nodes. First, a transformation module is used to enable end-to-end training of existing methods based on VGAE. Second, the pseudo node classification task is introduced into the network through multitask learning to make classification decisions for graph nodes. The graph nodes that have consistent decisions on clustering and pseudo node classification are added to a pseudo-label set, which can provide fruitful self-supervision for subsequent training. This pseudo-label set is gradually augmented during training, thus reinforcing the generalization capability of the network. Finally, we investigate different sorting strategies to further improve the quality of the pseudo-label set. Extensive experiments on multiple datasets show that the proposed method achieves outstanding performance compared with state-of-the-art methods. Our code is available at https://github.com/Jillian555/TNNLS_CDRS.

On the possibility of detection and classification of noise sources based on analysis of their trajectories at the output of adaptive spatial processing

To detect and classify objects, the source trajectories detected in the process of hydroacoustic observation are used, which contain information about the measured parameters of objects, which are their classification attributes. The analysis of these attributes allows to make a decision about the class of the observed object, for example, a surface or underwater source. As the measured object parameters are used their energy characteristics, parameters of the observed trajectory (bearing, speed of change in bearing and other possible trajectory parameters). In this case, the correctness and speed of the decision on classification depends on the quantity and quality of used classification signs, which are determined by both parameters of the observed object, and the features of sound propagation from the source to the observation device.For signal detection and resolution, we further consider fast projective adaptive algorithms, the use of which, as applied to the problems of in-situ experimental signal detection was considered in [1], [4]. The goal of this class of algorithms is to provide a high probability of detection and accurate measurement of the parameters of the source trajectories under the conditions of a model of multi-beam propagation and scattering in the real ocean environment [5]-[8]. The proposed work is a continuation of the work [1], [4], and aims to ensure the use of experimental field data not only for detection, but also for the classification of the observed sources.The subject of the study are the results of the full-scale experiment of hydroacoustic noise direction finding, given earlier and described in detail in [1], [4]. For the experiment was used antenna of L = 56 vertical daisy chains (of 10 elements each), equidistantly spaced horizontally. The antenna was installed at a depth of 200 meters in the coastal marine zone of the coastal wedge near the shipping lanes. The elements of the flat antenna were affected by signals from surface ships moving ncontrollably in the observation area and one underwater source.A singular decomposition of sampled antenna element data was used to construct adaptive algorithms. Modification of the initial results of singular decomposition allows to create algorithms that provide priority conditions for extraction of separate components of the observed (e. g., weakest) signals when constructing direction finding terrain.In this regard, in addition to the non-adaptive direction finding relief, it is proposed to form three variants of direction finding relief, each of which solves part of the general problem of selection and classification of individual varieties of observed signals:– Initial, corresponding to the energy of the signals of the input sample with amplified components of the weakest signals (overview algorithm);– direction finding relief, which uses an algorithm to detect weak and scattered signals;– direction finding terrain, which highlights the coherent components of signals.Analysis of the trajectories of more than 30 sources in the episode of two hours and forty minutes has been carried out, which increased the reliability of the detection and measurement accuracy of the parameters of the observed objects. Joint analysis of the trajectories of sources based on different variants of the bearing relief allowed to improve the conditions for the detection of weak signals and to make classification decisions using the classification attribute of the width of the area of fluctuations of the surface target trajectory for signals with a strong scattered component.

A Bootstrap-Aggregating in Random Forest Model for Classification of Corn Plant Diseases and Pests

Control of diseases and pests of maize plants is a significant challenge to ensure global food security, self-sufficiency, and sustainable agriculture. Classification or early detection of diseases and pests of corn plants is intended to assist the control process. Random forest is a classification model in tree-based statistical learning in making decisions. This approach is an ensemble method that generates many decision trees and makes classification decisions based on the majority of trees selecting the same class. However, tree-based methods are often unstable when small changes or disturbances exist in the learning data. Such instability can produce significant variances and affect model performance. This study classifies diseases and pests of the corn plant using a random forest method based on bootstrap-aggregating. It fits multiple models of a single random forest, then combines the predictions from all models and determines the final result using majority voting. The results showed that the bootstrap aggregating could improve the classification of diseases and pests of maize using a random forest if the number of trees is optimal.

Deep Learning-Based Intelligent Apple Variety Classification System and Model Interpretability Analysis

In this study, series networks (AlexNet and VGG-19) and directed acyclic graph (DAG) networks (ResNet-18, ResNet-50, and ResNet-101) with transfer learning were employed to identify and classify 13 classes of apples from 7439 images. Two training datasets, model evaluation metrics, and three visualization methods were used to objectively assess, compare, and interpret five Convolutional Neural Network (CNN)-based models. The results show that the dataset configuration had a significant impact on the classification results, as all models achieved over 96.1% accuracy on dataset A (training-to-testing = 2.4:1.0) compared to 89.4-93.9% accuracy on dataset B (training-to-testing = 1.0:3.7). VGG-19 achieved the highest accuracy of 100.0% on dataset A and 93.9% on dataset B. Moreover, for networks of the same framework, the model size, accuracy, and training and testing times increased as the model depth (number of layers) increased. Furthermore, feature visualization, strongest activations, and local interpretable model-agnostic explanations techniques were used to show the understanding of apple images by different trained models, as well as to reveal how and why the models make classification decisions. These results improve the interpretability and credibility of CNN-based models, which provides guidance for future applications of deep learning methods in agriculture.

Development of a k-Nearest Neighbor (kNN) Machine Learning Model to Estimate the SPT N-Values of Valenzuela City, Philippines

Due to the intricate geological formation of geomaterials, they often exhibit a range of attributes at different sites on a given site. This has posed a problem for geotechnical engineers, as they require correct soil and rock information to plan and design geotechnical construction projects. Numerous efforts have been made at the local level to bridge this disparity by standardizing the quantification of soil parameters; nevertheless, these studies have limitations, and there are still areas with ambiguous or incorrect information. To help close this gap, this article will demonstrate an innovative technique for predicting representative Soil Penetration Test (SPT) N-values for each Valenzuela City Barangay/Zone using a k-Nearest Neighbor (k-NN) Machine Learning Model. Borehole data from the city of Valenzuela were collected and analyzed for this study. The k-NN Model’s input parameters are the borehole’s latitude, longitude, and depth, while the response parameters are the SPT N-Values. The k-NN technique was used to train the input data, identify patterns, and make classification decisions for the SPT N-Values per depth per location in Valenzuela City. The centroids of each Barangay/Zone in Valenzuela City were also extracted in Latitude and Longitude format and utilized to deploy the k-NN Model. To illustrate the findings, we presented subsurface information from Valenzuela City with SPT N-Values. For validation purposes, the accuracy rate of the machine learning model was obtained. The model’s hyperparameter in terms of k was tuned to determine if the accuracy rate could be improved.

Taking Care of the Discretization Problem: A Comprehensive Study of the Discretization Problem and a Black-Box Adversarial Attack in Discrete Integer Domain

Neural network based (NN-based) classifiers are known vulnerable against adversarial examples, namely, adding slight perturbations to a benign image cause a classifier to make a false prediction. To evaluate the robustness of NN-based classifiers against adversarial examples, numerous adversarial attacks with high success rates have been proposed recently. NN-based image classifiers usually normalize valid images (e.g., RGB image where the value at each coordinate is an integer between 0 and 255) into a real continuous domain (e.g., 3-dimensional matrix where the value at each coordinate is a real number between 0 and 1) and make classification decisions on the normalized images. However, adversarial examples crafted in a real continuous domain may become benign once they are denormalized back into the corresponding discrete integer domain, known as the discretization problem. This problem has been mentioned in some prior works but received relatively limited attention. In this work, we report the first comprehensive study of existing works to understand the impacts of the discretization problem. By analyzing 35 representative methods and empirically studying 20 representative open source tools, we found 29/35 (theoretically) and 14/20 (empirically) are affected by the discretization problem, e.g., the success rate could dramatically drop from 100 to 10 percent after the domain transformation. As the first step towards addressing this problem in a black-box scenario, we propose a novel derivative-free optimization method, which can directly craft adversarial examples in the discrete integer domain. Experimental results show that the method achieves nearly 100 percent attack success rates for both targeted and untargeted attacks, comparable to the most popular white-box methods (FGSM, BIM and C&W), and significantly outperforms representative black-box methods (ZOO, AutoZOOM, NES-PGD, Bandits, FD, FD-PSO and GenAttack). Our results suggest that the discretization problem should be treated more seriously, and the discrete optimization algorithms show a promising future in crafting effective black-box attacks.

Recall Network: A Simple Brain-Inspired Algorithm for Classification.

The latest development of neuroscience has deepened the understanding of the information-processing mechanisms in the human brain and inspired a couple of sophisticated computational methods, such as deep learning, memory networks, and hierarchical temporal memory. However, it remains a challenge to explore simpler models due to the high computational cost of the above-mentioned methods. This paper proposes recall network (RN), an intuitive and simple model, that initializes itself by constructing the network path derived from the correlation of features in the training dataset and then makes classification decisions by recalling the paths that are relevant to the features in the test set. The algorithm has been applied to 263 datasets available from UCI Machine Learning Repository, and the classification results of repeated 10-fold cross-validation experiments on Weka demonstrate its competitive performance with prestigious classification algorithms, such as ANN, J48, and KNN.

ModelWise: Interactive Model Comparison for Model Diagnosis, Improvement and Selection

AbstractModel comparison is an important process to facilitate model diagnosis, improvement, and selection when multiple models are developed for a classification task. It involves careful comparison concerning model performance and interpretation. Current visual analytics solutions often ignore the feature selection process. They either do not support detailed analysis of multiple multi‐class classifiers or rely on feature analysis alone to interpret model results. Understanding how different models make classification decisions, especially classification disagreements of the same instances, requires a deeper model understanding. We present ModelWise, a visual analytics method to compare multiple multi‐class classifiers in terms of model performance, feature space, and model explanation. ModelWise adapts visualizations with rich interactions to support multiple workflows to achieve model diagnosis, improvement, and selection. It considers feature subspaces generated for use in different models and improves model understanding by model explanation. We demonstrate the usability of ModelWise with two case studies, one with a small exemplar dataset and another developed with a machine learning expert with real‐world perioperative data.

Types, characteristics and application of termination rules in computerized classification testing

<p id="p00005">Computerized classification testing (CCT) can adaptively classify test-takers into two or more different categories, and it has been widely used in qualifying tests and clinical psychology or medical diagnosis. As an essential part of CCT, the termination rule determines when the test is to be stopped and to which category the test-taker is ultimately classified into, directly affecting the test efficiency and classification accuracy. According to the theoretical basis of the termination rules, existing rules can be roughly divided into the likelihood ratio, Bayesian decision theory, and confidence interval rules. And their core ideas are constructing hypothesis tests, designing loss functions, and comparing the relative positions of confidence intervals, respectively. At the same time, when constructing specific termination rules, the requirement of different test scenarios (e.g., the number of categories and the number of tests’ dimensions) should also be considered. <br/>There are advantages and disadvantages to each of the three types of termination rules. Specifically, the likelihood ratio rule is based on the likelihood ratio test, with better theoretical properties. However, the method requires prior determination of the indifference interval and the type I and II error rates, introducing the impact of subjective factors. Also, it is more challenging to extend the method in complex test situations, such as multidimensional and multicategory CCT. Bayesian decision theory rules make classification decisions based on the loss function. It can dynamically optimize the decision from a more global perspective since it works backward from the final stage of the test. In addition, the variety of loss functions makes the method very flexible in form and makes it easy to be applied to different test situations. However, in practice, the flexibility will inevitably result in the uncertainty of the choice of loss function, and the inappropriate loss function may be biased. The confidence interval method is the most straightforward because of its relatively simple principle and low computational effort. However, this method is less robust and has a relatively low test efficiency. <br/>Currently, CCT is mainly applied in eligibility tests and clinical medicine questionnaires. In eligibility tests, all three types of termination rules have the potential to be widely applied. However, in practice, the principles of the likelihood ratio rule and the Bayesian decision theory rule are not easily understood by the general public, and these methods are also accompanied by the problem of over-exposure of items for their preference of cut-point based item selection methods. Therefore, the confidence interval rule, which is relatively simple in principle and has alleviated item exposure, has been widely used in existing qualifying tests. Bayesian decision theory rules are more applicable in clinical questionnaires because of their finer control over various classification losses. <br/>The following can be considered for future research on CCT termination rules. First, Bayesian decision theory rules can be improved by considering non-statistical constraints with the help of the flexibility of its loss function. Second, termination rules can be developed for multidimensional and multicategory CCT to meet more practical needs. Third, termination rules that integrate response time can be developed to improve test efficiency and classification accuracy. Fourth, it is possible to construct termination rules under the framework of machine learning.

A Machine Learning–Enabled Partially Observable Markov Decision Process Framework for Early Sepsis Prediction

Sepsis is a life-threatening condition, caused by the body’s extreme response to an infection. In the United States, 1.7 million cases of sepsis occur annually, resulting in 265,000 deaths. Delayed diagnosis and treatment are associated with higher mortality rates. An exponential rise in the availability of medical data has allowed for the development of sophisticated machine learning algorithms to predict sepsis earlier than the onset. However, these models often underperform, as the training data are retrospective and do not fully capture the uncertain future. In this study, we develop a novel framework, which we refer to as MLePOMDP, to leverage and combine the underlying, high-level knowledge about sepsis progression and machine learning (ML) for classification. Specifically, we use a hidden Markov model to describe sepsis development at a high level, where the ML model makes the higher-order “observations” from temporal data. Consequently, a partially observable Markov decision process (POMDP) model is developed to make classification decisions. We analytically establish that the optimal policy is of threshold-type, which we exploit to efficiently optimize MLePOMDP. MLePOMDP is calibrated and tested using high-frequency physiological data collected from bedside monitors. Different from past POMDP-based frameworks, MLePOMDP is developed for a prediction task using a very small state definition, produces highly interpretable results, and accounts for a novel and clinically meaningful action space. Our results show that MLePOMDP outperforms machine learning–based benchmarks by up to 8% in precision. Importantly, MLePOMDP is able to reduce false alarms by up to 28%. An additional experiment is conducted to show the generalizability of MLePOMDP to different patient cohorts. Summary of Contribution: This study develops a novel real-time decision support framework for early sepsis prediction by integrating well-known machine learning models (random forest and neural networks) with a well-established sequential decision-making model, namely, a partially observable Markov decision process (POMDP). The structural properties of the optimal policy are further explored and a threshold-type structure is established, which is then leveraged to develop a customized algorithm to solve the problem more efficiently. The resulting framework demonstrates the benefit of applying POMDPs to augment machine learning outputs. Specifically, the framework results in the reduction of false alarms in sepsis predictions where decisions are made in real time, hence improving the overall prediction precision.

Information-Extreme Machine Learning of Wrist Prosthesis Control System Based on the Sparse Training Matrix

The article considers the problem of machine learning of a wrist prosthesis control system with a non-invasive biosignal reading system. The task is solved within the framework of information-extreme intelligent data analysis technology, which is based on maximizing the system’s information productivity in machine learning. The idea of information-extreme machine learning of the control system for recognition of electromyographic biosignals, as in artificial neural networks, consists in adapting the input information description to the maximum total probability of making correct classification decisions. However, unlike neuro-like structures, the proposed method was developed within a functional approach to modeling the cognitive processes of the natural intelligence of forming and making classification decisions. As a result, the proposed method acquires the properties of adaptability to the intersection of classes in the space of recognition features and flexibility when retraining the system due to the recognition class alphabet expansion. In addition, the decision rules constructed within the framework of the geometric approach are practically invariant to the multidimensionality of the space of recognition features. The difference between the developed method and the well-known methods of information-extreme machine learning is the use of a sparse training matrix, which allows for reducing the degree of intersection of recognition classes significantly. The optimization parameter of the input information description, the training dataset, is the quantization level of electromyographic biosignals. As an optimization criterion is considered the modified Kullback information measure. The proposed machine learning algorithm results are shown in the example of recognition of six finger movements and wrist.

Classification Accuracy of Multidimensional Tests: Quantifying the Impact of Noninvariance

ABSTRACT There has been tremendous growth in research on measurement invariance over the past two decades. However, given that psychological tests are commonly used for making classification decisions such as personnel selections or diagnoses, surprisingly, there has been little research on how noninvariance impacts classification accuracy. Millsap and Kwok previously proposed a selection accuracy framework for that purpose, which has been recently extended to categorical data. Their framework, however, only deals with classification using a unidimensional test. In contrast, classification in practice usually involves multidimensional tests (e.g., personality) or multiple tests, with different weights assigned to each dimension. In the current paper, we extend Millsap and Kwok’s framework for examining the impact of noninvariance to a multidimensional test on classification. We also provide an R script for the proposed method and illustrate it with a personnel selection example using data from a published report featuring a five-factor personality inventory.

Exploration on the Teaching Reform Measure for Machine Learning Course System of Artificial Intelligence Specialty

Due to the particularity of the artificial intelligence major and the machine learning courses learned, the traditional course teaching model is not suitable for artificial intelligence major machine learning courses. Based on this background, this article proposes a new system based on machine learning curriculum teaching reform. It mainly includes the reform of curriculum teaching mode, curriculum practice reform, and teaching process reform. In order to verify the effect of the proposed new model on the teaching quality of machine learning courses, this article also proposes an evaluation method based on intelligent technology. Firstly, the feasibility of evaluation based on intelligent technology is described. Secondly, it lists the application details of the existing teaching evaluation based on intelligent technology. Finally, a novel teaching quality evaluation system based on intelligent technology is proposed. The system collects student facial expression data and uses classification algorithms to make classification decisions on the data. The result of the decision can give feedback on the quality of classroom teaching. The comparison of experiments based on different intelligent technologies shows that the teaching quality evaluation system proposed in this article is feasible and effective.

Detection Method for Classifying Malicious Firmware

A malicious firmware update may prove devastating to the embedded devices both that make up the Internet of Things (IoT) and that typically lack the same security verifications now applied to full operating systems. This work converts the binary headers of 40,000 firmware examples from bytes into 1024-pixel thumbnail images to train a deep neural network. The aim is to distinguish benign and malicious variants using modern deep learning methods without needing detailed functional or forensic analysis tools. One outcome of this image conversion enables contact with the vast machine learning literature already applied to handle digit recognition (MNIST). Another result indicates that greater than 90% accurate classifications prove possible using image-based convolutional neural networks (CNN) when combined with transfer learning methods. The envisioned CNN application would intercept firmware updates before their distribution to IoT networks and score their likelihood of containing malicious variants. To explain how the model makes classification decisions, the research applies traditional statistical methods such as both single and ensembles of decision trees with identifiable pixel or byte values that contribute the malicious or benign determination.

Assessing recovery of spectacled eiders using a Bayesian decision analysis.

Assessing species status and making classification decisions under the Endangered Species Act is a critical step towards effective species conservation. However, classification decisions are liable to two errors: i) failing to classify a species as threatened or endangered that should be classified (underprotection), or ii) classifying a species as threatened or endangered when it is not warranted (overprotection). Recent surveys indicate threatened spectacled eider populations are increasing in western Alaska, prompting the U.S. Fish and Wildlife Service to reconsider the federal listing status. There are multiple criteria set for assessing spectacled eider status, and here we focus on the abundance and decision analysis criteria. We estimated population metrics using state-space models for Alaskan breeding populations of spectacled eiders. We projected abundance over 50 years using posterior estimates of abundance and process variation to estimate the probability of quasi-extinction. The decision analysis maps the risk of quasi-extinction to the loss associated with making a misclassification error (i.e., underprotection) through a loss function. Our results indicate that the Yukon Kuskokwim Delta breeding population in western Alaska has met the recovery criteria but the Arctic Coastal Plain population in northern Alaska has not. The methods employed here provide an example of accounting for uncertainty and incorporating value judgements in such a way that the decision-makers may understand the risk of committing a misclassification error. Incorporating the abundance threshold and decision analysis in the reclassification criteria greatly increases the transparency and defensibility of the classification decision, a critical aspect for making effective decisions about species management and conservation.