Pattern Classification

Zero-day vulnerabilities remain one of the most critical challenges in modern cybersecurity due to their unknown nature and lack of available patches at the time of exploitation. Traditional signature-based intrusion detection systems fail to detect such threats effectively. Recent advances in machine learning (ML) and deep learning (DL) have enabled the development of adaptive intrusion detection systems capable of identifying unknown attack patterns. This paper presents a comprehensive survey of ML-based approaches for zero-day vulnerability detection, highlighting key methodologies, datasets, and evaluation frameworks. We further discuss experimental methodologies using benchmark datasets such as CICIDS2017 and UNSW-NB15, and present a reference architecture for MLbased zero-day detection systems.

The desire for persistent, long term surveillance and covertness places severe constraints on the power consumption of a sensor node. To achieve the desired endurance while minimizing the size of the node, it is imperative to use application-specific integrated circuits (ASICs) that deliver the required performance with maximal power efficiency while minimizing the amount of communication bandwidth needed. This paper reviews our ongoing effort to integrate several micropower devices for low-power wake-up detection, blind source separation and localization and pattern classification, and demonstrate the utility of the system in relevant surveillance applications. The capabilities of each module are presented in detail along with performance statistics measured during recent experiments.

Este trabajo solo persigue contribuir a describir e identificar problemas. Pero describir e identificar problemas para tratar de contribuir a alcanzar objetivos superiores y más abarcadores en las Ciencias Médicas. Problemas en los que con alguna frecuencia no nos detenemos, que a veces parecen algo casi terminado, casi acabado, casi perfecto. Problemas en los que con mucha frecuencia se piensa que no se necesita invertir mucho tiempo, pues ya casi se tiene todo lo que se necesita, y solo pequeños ajustes que algún selecto grupo de pensadores se ha de encargar de llevar a cabo cada vez que sea necesario. ¿Qué es la salud para la medicina científica? El concepto de Salud de la medicina occidental moderna (M.O.M.) ha atravesado por diversas proyecciones. Una, por ejemplo, la limitaba a la ausencia de enfermedad, respondiendo a un momento de auge del enfoque nosológico. Otra, posterior, decía que Salud no es solamente la ausencia de enfermedad, sino el completo bienestar físico, mental y social del hombre. En ésta, a más de manifestarse claramente su principio dicotómico también, se expresan algunas inquietudes sociales emergentes, en especial después de la Segunda Guerra Mundial, y una incipiente influencia de la necesidad del concepto de calidad de vida en el momento de su formulación. Sin embargo, este segundo concepto tiene también limitaciones importantes. El completo bienestar físico, mental y social del hombre no es una utopía, sino que es un objetivo imposible de alcanzar siquiera teóricamente, pues la solución de un problema o de una necesidad implica siempre, por lo menos, la generación de otro. Pongamos un ejemplo que no nos saque del tema. La reducción de la mortalidad infantil es una tarea impostergable para la humanidad desde muchos puntos de vista. Resulta vergonzosa la displicencia con que muchos gobiernos han tratado la solución de este tema o la "naturalidad", por no decir indiferencia con que muchos contemplan como cientos de miles de niños mueren semanalmente en el mundo por motivos perfectamente evitables o de fácil solución. No parece que seamos la clase de seres civilizados de la que no pocas veces nos ufanamos. Sin embargo, al cumplir con esta elemental obligación humana, ética y moral, estamos contribuyendo a crear un problema mayor. Fue Carlos Darwin el que llamó "Selección Natural" a ese proceso de constante fortalecimiento y perfeccionamiento de las especies mediante la eliminación temprana de los menos aptos. Al violentar este proceso, estamos contribuyendo a que nuestra especie sea cada vez más frágil y con una menor capacidad de adaptación ante la diversidad de contingencias que debe afrontar un ser vivo. Si bien es cierto que no es solo el incremento de la supervivencia de los niños el único factor que contribuye a disminuir la eficiencia adaptativa de nuestra especie, no es menos cierto que juega un papel de consideración. Es indispensable resolver cada vez más y mejor el problema de la mortalidad infantil, pero es también de extrema prioridad comenzar a estudiar y resolver con cada vez mayor profundidad y precisión el problema que estamos contribuyendo a crear. Resulta que, al resolver un problema, estamos propiciando el surgimiento otro de mayor magnitud indiscutiblemente, en tanto abarcará a toda la especie. Además de sus aristas de quimérica ficción, esta segunda definición de salud resulta un concepto tan amplio y difuso que, con las herramientas de la M.O.M., no es factible operar con él. La M.O.M. proclama que "no existen enfermedades, sino enfermos", lo que con frecuencia sirve para justificar dos verdades inconmovibles: a) que todas las enfermedades no se expresan ni evolucionan igual en todos los pacientes b) que todos los enfermos no responden igual a los mismos tratamientos. Llama la atención cómo esta afirmación coexiste en armonía con un concepto de salud que comienza por negarlo en su primera frase: "no es solo la ausencia de enfermedad". Esta armónica convivencia contribuye a denunciar que con disimulo, de manera un poco subrepticia, la M.O.M. no opera con enfermos sino con enfermedades. Es por esa razón que el sinónimo de enfermedad pueda ser "entidad nosológica 1 ". También asombra con qué facilidad tantas veces se afirma que una enfermedad no tiene nada que ver con otra u otras, a pesar de que han ocurrido en un mismo individuo a lo largo de su vida o coincidiendo en un tiempo aproximado. Si no hay enfermedades, sino enfermos, ¿qué sentido tiene hablar de salud y enfermedad? Es evidente que se proclama que no hay enfermedades sino enfermos, pero no se opera con enfermos, sino con enfermedades. Para la metafísica, un fenómeno existe o no existe, como tampoco puede ser lo que es y, al mismo tiempo, algo distinto. Lo positivo y lo negativo se excluyen, revisten la forma de una antítesis rígida. A primera vista este método discursivo pudiera parecer razonable para algunos, y pudiera resultar incluso de utilidad práctica como parte del proceso de determinadas zonas del pensamiento dependiendo de la naturaleza del objeto de su estudio, pero termina por tropezar, con las cualidades de un método parcial, limitado que, absorbido por los fenómenos concretos, no alcanza a ver su concatenación; concentrado en su estatismo, no alcanza a ver su dinámica i . Salud y enfermedad se excluyen, se contraponen. O se está sano o se está enfermo. ¿Es admisible esta disyuntiva o es una proposición metafísica? ¿No recuerda la contraposición de sano y enfermo a la de vivo o muerto o a la de verdad y error? La salud en la medicina china tradicional.

Abstract Automatic pattern classification is a very important field of artificial intelligence. For these kind of tasks different techniques have been used. In this work a combination of decision trees and self-organizing neural networks is presented as an alternative to attack the problem. For the construction of these trees growth processes are applied. In these processes, the evaluation of classification efficiency of one or several nodes in different configurations is necessary in order to take decisions to optimize the structure and performance of the self-organizing neural tree net. In order to perform this task a group of coefficients that quantify the efficiency is defined and a growth algorithm based on these coefficients is developed. In the tests, a comparison with other classification methods, using cross-validation methods with real and artificial databases, is carried out.

Constructive learning algorithms o er an approach to incremental construction of near-minimal arti cial neural networks for pattern classi cation. Examples of such algorithms include Tower, Pyramid, Upstart, and Tiling algorithms which construct multilayer networks of threshold logic units or, multilayer perceptrons. These algorithms di er in terms of the topology of the networks that they construct which in turn biases the search for a decision boundary that correctly classi es the training set. This paper presents an analysis of such ...

This paper aims to separate different snow regions over the terrestrial ice sheets based on their measured microwave signatures. It takes advantage of coregistered data from passive and active sensors on the Environmental Satellite (Envisat) to directly derive a snow facies indicator in a point-by-point basis. This paper represents the first attempt of this kind in exploiting nadir-viewing and dual-frequency data from both altimeter and radiometer sensors. The approach is based on a clustering method. Such representation of the data by means of fewer clusters necessarily loses fine details but achieves simplification in geographical representation and eases the description of the condition of the ice sheets in 2004. Our approach broadens the description of the snow pack by taking into account characteristics such as surface roughness, grain size, stratification, and snowmelt effects, whereas the latter has often solely been considered in most previous work. Such partition of the ice sheets might help to better understand relationships between microwave signatures and snow morphology. It could also be useful for estimating elevation uncertainty in altimeter data, which, in turn, is essential to correctly interpret the significance of the rates of elevation change in a changing climate and to convert elevation change to snow mass change.

This article examines the emerging role of artificial intelligence, predictive analytics, and neuroscientific forecasting in reshaping contemporary understandings of criminal liability. As criminal justice systems increasingly integrate algorithmic risk assessment, behavioral prediction models, and brain-based forecasting tools, traditional doctrines of mens rea and individualized culpability face unprecedented epistemic challenges. The study argues that predictive systems do not replace normative legal judgment but create a new evidentiary layer that influences attribution of dangerousness, intent inference, sentencing calibration, and preventive detention. A doctrinal framework of algorithmic culpability is proposed to reconcile predictive technologies with due process, evidentiary reliability, and constitutional safeguards. https://doi.org/10.5281/zenodo.19647127

The sunspot classification scheme presented in this paper is considered as a 2-D classification problem on archived datasets, and is not a real-time system. As a first step, it mirrors the Zürich/McIntosh historical classification system and reproduces classification of sunspot patterns based on preprocessing and neural net training datasets. Ultimately, the project intends to move from more rudimentary schemes, to develop spatial-temporal-spectral classes derived by correlating spatial and temporal variations in various wavelengths to the brightness fluctuation spectrum of the sun in those wavelengths. Once the approach is generalized, then the focus will naturally move from a 2-D to an n-D classification, where 'n' includes time and frequency. Here, the 2-D perspective refers both to the actual SOHO Michelson Doppler Imager (MDI) images that are processed, but also refers to the fact that a 2-D matrix is created from each image during preprocessing. The 2-D matrix is the result of running Principal Component Analysis (PCA) over the selected dataset images, and the resulting matrices and their eigenvalues are the objects that are stored in a database, classified, and compared. These matrices are indexed according to the standard McIntosh classification scheme. The initial dataset, provided by LMSAL, consisted of 1596 MDI images recorded during the period 05 March 2001 to 26 December 2004, representing 20 different McIntosh classes of sunspots. To create consistent training sets and matching schemes, images were first selected that were 183x183 pixels square (3 arcmin square); non-square images were rejected in the first pass. In addition, the color resolution factors in the data carried either 8-bit mapping [0-255] or full color resolution, scaled at each image. The full color resolution images were selected in preference to the 8-bit, and max-min analysis was carried out so that all the remaining images could be normalized. Solar limb images were purged; and noisy data were eliminated for this first pass. The concept 'noisy' refers to the fact that some images have a variation in background brightness due to, for example, off-center image angles. Such variation in scene brightness could confound the neural net during this initial training development, so those images were also purged from the initial test dataset. A next step would be to include a threshold determination for each image based on contrast between sunspot pattern and local background. One might eventually include a quantitative noise contrast based on variations in the full color resolution scaling for a particular image. However, it is also worth noting that images with more than one sunspot class type in the same image were kept. The result of this filtering left 308 images, now representing 15 different McIntosh classes. Images for a training dataset were selected purely randomly, but all 15 classes were included.

Digital mammography is one of the most suitable methods for early detection of breast cancer. It uses digital mammograms to find suspicious areas containing benign and malignant microcalcifications. However, it is very difficult to distinguish benign and malignant microcalcifications. This is reflected in the high percentage of unnecessary biopsies that are performed and many deaths caused by late detection or misdiagnosis. A computer based feature selection and classification system can provide a second opinion to the radiologists in assessment of microcalcifications. The research in this paper proposes and investigates a neural-genetic algorithm for feature selection in conjunction with neural and statistical classifiers to classify microcalcification patterns in digital mammograms. The obtained results show that the proposed approach is able to find an appropriate feature subset and neural classifier achieves better results than two statistical models.

A novel encoding technique is proposed for the recognition of patterns using four different techniques for training artificial neural networks (ANNs) of the Kohonen type. Each template or model pattern is overlaid on a radial grid of appropriate size, and converted to a two-dimensional feature array which then acts as the training input to the ANN. The first technique employs Kohonen's self-organizing network, each neuron of which is assigned, after training, the label of the model pattern. It is found that a graphical plot of the labels of the neurons exhibits clusters (which means in effect that the feature array pertaining to distorted versions of the same pattern belongs to a specific cluster), thereby justifying the coding strategy used in this paper. When the new, unknown pattern is input to the network, it is classified to have the same label of the neuron whose corresponding model pattern is closest to the given pattern. In an attempt to reduce the computational time and the size of the network, and simultaneously improve accuracy in recognition, Kohonen's learning vector quantization (LVQ) algorithm is used to train the ANN. To further improve the network's performance and to realize a network of minimum size, two constructive learning algorithms, both based on LVQ, are proposed: (1) multi-step learning vector quantization (MLVQ), and (2) thermal multi-step learning vector quantization (TLVQ). When the proposed algorithms are applied to the classification of noiseless and noisy (and distorted) patterns, the results demonstrate that the pattern encoding strategy and the suggested training techniques for ANNs are efficient and robust. For lack of space, only the most essential results are presented here. For details, see Ganesh .

A limiting factor for the application of IDA methods in many domains is the incompleteness of data repositories. Many records have fields that are not filled in, especially, when data entry is manual. In addition, a significant fraction of the entries can be erroneous and there may be no alternative but to discard these records. But every cell in a database is not an independent datum. Statistical relationships will constrain and, often determine, missing values. Data imputation, the filling in of missing values for partially missing data, can thus be an invaluable first step in many IDA projects. New imputation methods that can handle the large-scale problems and large-scale sparsity of industrial databases are needed. To illustrate the incomplete database problem, we analyze one database with instrumentation maintenance and test records for an industrial process. Despite regulatory requirements for process data collection, this database is less than 50% complete. Next, we discuss possible solutions to the missing data problem. Several approaches to imputation are noted and classified into two categories: data-driven and model-based. We then describe two machine-learning-based approaches that we have worked with. These build upon well-known algorithms: AutoClass and C4.5. Several experiments are designed, all using the maintenance database as a common test-bed but with various data splits and algorithmic variations. Results are generally positive with up to 80% accuracies of imputation. We conclude the paper by outlining some considerations in selecting imputation methods, and by discussing applications of data imputation for intelligent data analysis.

Un sistema filosofico exhaustivo de 37 capitulos y 8 bloques tematicos para comprender y transformar realidades complejas. Incluye el Verificador Dialectico: primera operacionalizacion computacional asistida por IA del metodo dialectico. Edicion expandida con ejemplificaciones, casos de estudio y analisis comparativos.

alerts, prioritize response, and audit model behavior. At the same time, the data conditions under which IDS models are trained are often unfavorable: labeled cybersecurity datasets are frequently small due to the cost of expert labeling and the rapid Explainable AI (XAI) is increasingly required for intrusion detection systems (IDS) because security analysts must justify alerts, prioritize response, and audit model behavior. In operational environments, however, supervised IDS commonly faces two constraints: limited labeled training data and imperfect supervision arising from delayed ground truth and weak labeling pipelines. This study presents a comparative evaluation of explainable multi-class intrusion detection under controlled small-data and noisy-label regimes using UNSW-NB15 and CICIDS2017. We simulate data scarcity by stratified downsampling of the training set and simulate label noise using both symmetric corruption and a security-realistic benignification mechanism that preferentially flips attack labels toward benign. Representative detector families are trained using empirical risk minimization and noise-mitigation strategies, and explanations are generated using SHAP, LIME, and Integrated Gradients. The evaluation jointly considers detection effectiveness, probability reliability, and explanation quality using Macro-F1, AUROC, AUPRC, Expected Calibration Error, and explanation metrics that capture faithfulness, stability, sparsity, and drift. Results show that performance and explanation reliability degrade nonlinearly when small data and noisy labels co-occur, with benignification noise causing the most severe losses. Noise-tolerant learning reduces these losses and improves calibration and explanation stability, indicating that training choices affect not only accuracy but also the reliability of analyst-facing explanations under scarce and noisy supervision.

The synthesised findings of five recent studies on the implementation of deep learning (DL) models to achieve adaptive malware classification and behavioural threat analysis in dynamic cyber environments are presented in this empirical review. The reviewed research shows that DL methods, convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformers, and hybrid approaches are significantly more efficient tools in terms of accuracy, feature extraction, and adaptation to the complexity of malware behaviors than traditional machine learning methods. Hybrid models that merge supervised, unsupervised, and federated learning are even better at improving the scalability and privacy preservation. Despite these developments, several challenges persist, including high computational costs, limited access to high-quality labeled data, susceptibility to adversarial attacks, the need for explainability, and the incorporation of these systems into real-world settings. Research gaps were noted in creating lightweight IoT and edge computing systems, adversarially robust DL models, explainable AI, and continual learning systems. The article review generally confirms the instrumental role of DL in developing adaptive cybersecurity strategies, emphasizing the need for innovative, interdisciplinary, and practical implementations that match the sophistication and practicality required in a dynamic digital environment.

Fake currency, unauthorized imitation money lacking government approval, constitutes a form of fraud. Particularly in Afghanistan, the prevalence of fake currency poses significant challenges and detrimentally impacts the economy. While banks and commercial establishments employ authentication machines, the public lacks access to such systems, necessitating a program that can detect counterfeit banknotes accessible to all. This paper introduces a method using image processing to identify counterfeit Afghan banknotes by analyzing specific security features. Extracting first and second order statistical features from input images, the WEKA machine learning tool was employed to construct models and perform classification with Random Forest, PART, and Naïve Bayes algorithms. The Random Forest algorithm achieved exceptional accuracy of 99% in detecting fake Afghan banknotes, indicating the efficacy of the proposed method as a solution for identifying counterfeit currency.

This paper is concerned with the development of Back-propagation Neural Network for Bangla Speech Recognition. In this paper, ten bangla digits were recorded from ten speakers and have been recognized. The features of these speech digits were extracted by the method of Mel Frequency Cepstral Coefficient (MFCC) analysis. The mfcc features of five speakers were used to train the network with Back propagation algorithm. The mfcc features of ten bangla digit speeches, from 0 to 9, of another five speakers were used to test the system. All the methods and algorithms used in this research were implemented using the features of Turbo C and C++ languages. From our investigation it is seen that the developed system can successfully encode and analyze the mfcc features of the speech signal to recognition. The developed system achieved recognition rate about 96.332% for known speakers (i.e., speaker dependent) and 92% for unknown speakers (i.e., speaker independent).

In many machine learning applications, one has access, not only to training data, but also to some high-level a priori knowledge about the desired behavior of the system. For example, it is known in advance that the output of a character recognizer should be invariant with respect to small spatial distortions of the input images (translations, rotations, scale changes, etcetera). We have implemented a scheme that allows a network to learn the derivative of its outputs with respect to distortion operators of our choosing. This not only reduces the learning time and the amount of training data, but also provides a powerful language for specifying what generalizations we wish the network to perform.

Memory-based classification algorithms such as radial basis functions or K-nearest neighbors typically rely on simple distances (Euclidean, dot product ... ), which are not particularly meaningful on pattern vectors. More complex, better suited distance measures are often expensive and rather ad-hoc (elastic matching, deformable templates). We propose a new distance measure which (a) can be made locally invariant to any set of transformations of the input and (b) can be computed efficiently. We tested the method on large handwritten character databases provided by the Post Office and the NIST. Using invariances with respect to translation, rotation, scaling, shearing and line thickness, the method consistently outperformed all other systems tested on the same databases.

Gait based gender identification has received a great attention from researchers in the last decade due to its potential in different applications. This will help a human identification system to focus only on the identified gender related features, which can improve search speed and efficiency of the retrieval system by limiting the subsequent searching space into either a male database or female database. In this paper after preprocessing, four binary moment features and two spatial features are extracted from human silhouette. Then the extracted features are used for training and testing two different pattern classifiers kNearest Neighbor (kNN) and Support Vector Machine(SVM). Experimental results show superior performance of our approach to the existing gender classifiers. To evaluate the performance of the proposed algorithm experiments have been conducted on NLPR database.

In the nearest-convex-model type classifiers, each class in the training set is approximated with a convex class model, and a test sample is assigned to a class based on the shortest distance from the test sample to these class models. In this paper, we propose new methods for approximating the distances from test samples to the convex regions spanned by training samples of classes. To this end, we approximate each class region with a polyhedral convex conic region by utilizing polyhedral conic functions (PCFs) and its extension, extended PCFs. Then, we derive the necessary formulations for computing the distances from test samples to these new models. We tested the proposed methods on different high-dimensional classification tasks including face, digit, and generic object classification as well as on some lower-dimensional classification problems. The experimental results on different datasets show that the proposed classifiers achieve either the best or comparable results on high-dimensional classification problems compared to other nearest-convex-model classifiers, which shows the superiority of the proposed methods.

This paper presents a novel deep learning-based travel behaviour choice model. Our proposed Residual Logit (ResLogit) model formulation seamlessly integrates a Deep Neural Network (DNN) architecture into a multinomial logit model. Recently, DNN models such as the Multi-layer Perceptron (MLP) and the Recurrent Neural Network (RNN) have shown remarkable success in modelling complex and noisy behavioural data. However, econometric studies have argued that machine learning techniques are a 'black-box' and difficult to interpret for use in the choice analysis. We develop a data-driven choice model that extends the systematic utility function to incorporate non-linear cross-effects using a series of residual layers and using skipped connections to handle model identifiability in estimating a large number of parameters. The model structure accounts for cross-effects and choice heterogeneity arising from substitution, interactions with non-chosen alternatives and other effects in a non-linear manner. We describe the formulation, model estimation, interpretability and examine the relative performance and econometric implications of our proposed model. We present an illustrative example of the model on a classic red/blue bus choice scenario example. For a real-world application, we use a travel mode choice dataset to analyze the model characteristics compared to traditional neural networks and Logit formulations. Our findings show that our ResLogit approach significantly outperforms MLP models while providing similar interpretability as a Multinomial Logit model.