The Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM) is a natural Bayesian nonparametric extension of the classical Hidden Markov Model for learning from (spatio-)temporal data. A sticky HDP-HMM has been proposed to strengthen the self-persistence probability in the HDP-HMM. Then, disentangled sticky HDP-HMM has been proposed to disentangle the strength of the self-persistence prior and transition prior. However, the sticky HDP-HMM assumes that the self-persistence probability is stationary, limiting its expressiveness. Here, we build on previous work on sticky HDP-HMM and disentangled sticky HDP-HMM, developing a more general model: the recurrent sticky HDP-HMM (RS-HDP-HMM). We develop a novel Gibbs sampling strategy for efficient inference in this model. We show that RS-HDP-HMM outperforms disentangled sticky HDP-HMM, sticky HDP-HMM, and HDP-HMM in both synthetic and real data segmentation.

Fiducial markers play a crucial role in localization, navigation, and tracking systems, particularly in robotics. However, widely used marker detection methods relying on RGB cameras often fail in low-light or no-light environments, where traditional black-and-white markers become ineffective. To address this challenge, we propose a Variable Lighting Conditions Fiducial Marker (VLCFM) system that enhances detection accuracy across diverse lighting conditions, ensuring reliable performance in complete darkness while remaining undetectable to the human eye.

The field of Artificial Intelligence (AI)-based weather forecasting is growing rapidly, with continuous progress in model development, techniques, and performance improvements. This paper provides a comprehensive overview of AI-based weather forecasting models, focusing on their current status, challenges, and directions for further development. A review of more than 40 models, primarily proposed after 2015, underscores the importance of critically examining various aspects of AI-based forecasting. Unlike previous reviews that targeted only a limited number of models or features, this study addresses a complete set of aspects and analyzes existing challenges from multiple perspectives. These aspects include the Machine Learning (ML) and Deep Learning (DL) methods used, datasets, predictand parameters, overfitting, and capability for forecasting extreme weather, lead time, spatiotemporal scale, performance criteria, overfitting, data assimilation, data-driven models, and the analysis of state-of-the-art (SOTA) models such as FengWu, ClimaX, Pangu-Weather, FourCastNet, GraphCast, GenCast, and Artificial Intelligence Forecasting System (AIFS) from various viewpoints. The review also discusses current challenges, including limited historical data and data quality, small-scale weather forecasting, model explainability, uncertainty, extreme weather prediction, physical constraints, temporal adaptation, and generalization, and outlines potential future directions.

This paper concerns improving recommender systems using evolutionary algorithms that optimize the composition of the directory of products in the recommender system. It focuses on merging some infrequent products into certain groups of similar products and replacing the regular products by some aggregates of products, such as a category of products or a cluster of products in the product embedding space, and preparing recommendations with such aggregates. Evolutionary algorithms endeavor to decide which products should be processed as single products and which ones should be merged with others into smaller or larger groups of products (merging products into groups is done by a given auxiliary merging strategy that cluster similar products into groups) with the aim to maximize the accuracy of the recommender system. Computational experiments on benchmark datasets confirmed the efficiency of the approach.

Abstract The Inventory Routing Problem (IRP) is a combinatorial optimization problem that combines routing decisions with inventory management. In this paper, an approach to solving the IRP is studied, which aims at using an external knowledge source (a known good solution or user interaction) to improve the results attained by an evolutionary algorithm solving an IRP instance. The proposed method improves the best solution found by the evolutionary algorithm by modifying schedules for some of the retailers according to those present in the known good solution or to schedules provided by a domain expert. The experiments shown that to improve the optimization results it suffices to perform a few repetitions of the knowledge import procedure. This observation motivates further research on user-interactive optimization algorithms for the IRP, because the number of interactions needed to improve the results can easily be handled by the user.

This paper presents a novel convolutional neural network (CNN) architecture for image steganalysis. We trained the CNN to detect distortions introduced by the state-of-the-art embedding algorithms: WOW, HUGO, and UNIWARD. We used a large database of test images and a large number of stegomessages, which resulted in a six billion-sample training set. The experimental results show that the resultant CNN, which was trained on such a large dataset, outperforms state-of-the-art steganalysis algorithms in terms of total error rate.

The Firefighter Problem (FFP) is an optimization problem of developing an optimal strategy for assigning firemen to nodes of a given graph in successive iterations of a simulation of spread of fires in the graph. This paper focusses on an extension of the original FFP, namely the Bi-Firefighter Problem (FFP2), where the second element (water) is introduced. FFP2 corresponds to the practical optimization problems, where more than one disease is spreading in the environment, and the objective is to minimize the total loss. Since the loss may come from two different sources, each of which causes different damages, the objective function is more complex than in the case of the original FFP. In this paper, an evolutionary approach to FFP2, the EA-FFP2 algorithm, based on a multi-modal Estimation of Distribution Algorithm (EDA), is proposed. EA-FFP2 was validated on a number of benchmark FFP2 instances that were also solved by the branch and bound algorithms or the heuristic local search algorithms run for a large number of iterations for a long time. Computational experiments confirmed that EA-FFP2 was capable of solving FFP2 and finding solutions close to the optima determined by the branch and bound algorithms or to the quasi-optima determined by exhaustive local search.

This paper concerns extracting knowledge from limit order books and using it to predict significant changes in the price of the financial asset under study. Limit order books are encoded in a feature-based data representation. A binary Support Vector Machine classifier is trained to predict whether a particular limit order book leads to a significant change in the price in a few successive time instants, or not. An Evolutionary Algorithm is used to find the optimal parameter setting of the feature-based data representation, so that the performance of the classifier was better. Computational experiments performed on financial ultra-high frequency time series coming from the London Stock Exchange Rebuild Order Book database confirmed that the optimization of the feature-based data representation is very effective in improving the performance of the classification.

This paper proposes an evolutionary approach to optimization of data representation for classification of patterns in financial ultra-high frequency time series. Input data describe order book shapes defined by sequences of price-capital pairs coming from ask and bid orders registered on the stock market. Target data describe the change of the mid price. Classifiers based on SVM try to predict the direction of changes of the mid price on the basis of the input data. An important problem in such an approach is the representation of the input data, because the raw input data consist of long and irregular queues of orders, and need to be transformed into a feature vector. An evolutionary approach is proposed to optimize the representation of the input data by gaussian curves. Experiments performed on real-world data from the London Stock Exchange Rebuilt Order Book database confirms that the evolutionary algorithm is capable of improving significantly the results of classification.

Next-generation sequencing allows high-throughput measurements of non-coding RNA expression levels in tissues. Analysis of microRNAs (miRNAs) is particularly effective in differentiation of cancerous tissue samples, based on patterns of their expression levels. The paper presents a wrapper feature selection approach based on t-Distributed Stochastic Neighbor Embedding (t-SNE), Covariance Matrix Adaptation Evolution Strategy (CMA-Es) and Support Vector Machine (SVM). The advantage of t-SNE is amplification of pairwise similarities by the means of t-Student neighborhood function. The attributes are embedded into 1-D space to reveal similarities between the features. Such information is used by CMA-ES through real-valued encoding in order to model pairwise relations between miRNAs with covariance matrices. Finally, the wrapper uses SVM to evaluate the objective, which expresses the tradeoff between classification quality and the desired number of features. The approach is tested on eight different cancer types from The Cancer Genome Atlas. It allows to find small sets of miRNAs to differentiate cancer types from a single tumor class to the normal one with high certainty.

This paper concerns knowledge discovery from ultra-high frequency financial time series of order book data from the London Stock Exchange Rebuild Order Book database. It focuses on extracting frequent patterns from order book shapes, which characterize particular stock market states. Because order book shapes vary considerably in time, it is necessary to convert the original price-volume representation to real vectors of a fixed length. In this paper two methods of representing order books are studied: using Gaussian Mixture Models (GMMs), and Gaussian density-based representation (GD). Both methods produce real vectors of a constant length and frequent patterns are defined as codebooks of clusters of these real vectors. To obtain useful knowledge we would like these patterns to represent different states of the market, be stable in time, and be universal. We define universality as periodical reoccurence, which makes historical knowledge usable for future decision making. Because of these time-related requirements a typical spatial clustering is not sufficient. This paper studies the usage of an evolutionary algorithm to obtain patterns with desirable properties.

Microarray technology allows to investigate gene expression levels by analyzing high dimensional datasets of few samples. Selection of discriminative, differentially expressed genes from such datasets is important to differentiate, prognose and understand the underlying biological processes. In this regard, the paper presents a new evolutionary gene selection method based on Student-t Stochastic Neighbor Embedding (t-SNE), Differential Evolution (DE) and Support Vector Machine (SVM). Here the underlying classification task of SVM is used as an optimization problem of DE, while t-SNE provides better ordering of genes for selection purpose. Generally, t-SNE is used to reorder the genes in such a way so that similar genes are grouped together and dissimilar genes are kept further apart. These reordered genes are then fragmented into fixed-length partitions. Thereafter, from each partition, a gene is selected randomly to encode the initial population of DE along with the combination of its weight and threshold values in order to participate in fitness computation. In the final generation of DE, a subset of genes is selected based on higher classification accuracy. The proposed technique is tested on six publicly available microarray datasets concerning various cancerous tissues of Homo sapiens and yields a potential set of genes by providing prefect or nearly perfect classification accuracy. Moreover, the superiority of the proposed technique has been demonstrated in comparison with other widely used techniques. Finally, the achieved results have also been justified by a statistical test and allowed us to draw biological conclusions through the identification of Gene Ontologies.

This paper presents two evolutionary algorithms, ECGA and BOA, applied to constructing stock market trading expertise, which is built on the basis of a set of specific trading rules analysing financial time series of recent price quotations. A few modifications of ECGA are proposed in order to reduce the computing time and make the algorithm applicable for real-time trading. In experiments carried out on real data from the Paris Stock Exchange, the algorithms were compared in terms of the efficiency in solving the optimization problem, in terms of the financial relevance of the investment strategies discovered as well as in terms of the computing time.

In this paper, a new functionality of early warning for an online stock trading system is presented. The warning functionality helps to focus traders' attention on specific situations on the stock market. The specific situations relate to the rare circumstances where a trader should be alerted by exceptional raises or drops of share prices, volatilities and market index changes. Usually, these alerts force a trader to make a decision either to buy or sell a share. To discover the warning rules and events, an evolution-based model is proposed. This model also introduces a new function that stores the experimental knowledge by keeping track of all historical alert events-solutions and actions taken by a trader. This model is composed of the three following components, which are integrated with each other: alert rules, pattern clustering and genetic engine. This approach has been tested on real data extracted from the Internet Bourse Expert System and Paris Stock Exchange.

This paper addresses the problem of constructing real-time stock trading expertise for financial time series. The expertise is arrived at via an evolutionary algorithm on the basis of a set of specified trading rules. As in most real-time expert systems, one of the main bottlenecks is the time constraint. In this paper, two approaches were compared using our system, Bourse-Expert, the first based on 350 trading rules, and the second based on 150 particular linear combinations of these 350 rules. Experiments carried out on real data from the Paris Stock Exchange showed that focusing on only 150 rules highly reduced the computation time without significantly reducing the quality of the expertise.