ABSTRACT. Phase stability testing and phase equilibrium calculations are central problems of thermodynamics of multicomponent mixtures with many applications in chemical engineering, enhanced oil recovery, or CO2 sequestration. The thermodynamical basis of both problems was formulated by J. W. Gibbs (~1885), and the first computational methods for solving these problems appeared at the very beginning of the digital computers era (Rachford and Rice, 1954). The conventional computational methods used until today originate in early 1980's. In the conventional approach, the mixture is described by the pressure, temperature, and overall chemical composition (PTN-formulation). The task is to determine whether the mixture remains stable or splits into two (or more) phases. In case of phase splitting, one has to determine the chemical compositions and amounts of the phases. The stable equilibrium state corresponds to the global minimum of the total Gibbs free energy of the system. These methods can be subsequently used as a part of a compositional transport simulation where the phase equilibrium calculation is typically performed at each finite element at every time step, requiring the computational methods to be both robust and efficient. The challenging part is the correct coupling of local thermodynamic equilibrium calculations with the compositional transport model. In this talk, I will show two alternative formulations of the basic problems where the state of the mixture is described either by volume, temperature, and moles (VTN-formulation), or internal energy, volume, and moles (UVN-formulation). Despite of the fact that these formulations were overlooked previously in the literature, I  will show that these formulations are in a sense more natural than the conventional approach when one wants to couple them with the transport simulation. I will present a unified approach that allows development of a single solver that can treat all three formulations in a unified way. I will also present examples of compositional simulations and phase equilibrium calculations in various formulations and recent results showing the extension of the conventional approach for calculation of phase equilibria of mixtures splitting up to four phases. 

ABSTRACT. As the quest for more sustainable and environmentally friendly materials has increased in the last decades, cellulose nanofibrils (CNFs), abundant in nature, have proven their capabilities as building blocks to create strong and stiff filaments. Experiments have been conducted to characterize CNFs with a rheo-optical flow-stop technique to study the Brownian dynamics through the CNFs' birefringence decay after stop. This paper aims to predict the initial relaxation of birefringence using Principal Component Analysis (PCA) and Long Short-Term Memory (LSTM) networks. By reducing the dimensionality of the data frame features, we can plot the principal components (PCs) that retain most of the information and treat them as time series. We employ LSTM by training with the data before the flow stops and predicting the behavior afterward. Consequently, we reconstruct the data frames from the obtained predictions and compare them to the original data.

ABSTRACT. Stance detection, i.e.\ recognition of utterances in favor, against or neutral in relation to some targets is important for text analysis. However, different approaches were tested on different datasets, often interpreted in different ways. We propose a unified overview of the state-of-the-art stance detection methods in which targets are expressed by short phrases. Special attention is given to zero-shot learning settings. An overview of the available multiple target datasets is presented that reveals several problems with the sets and their proper interpretation. Wherever possible, methods were re-run or even re-implemented to facilitate reliable comparison. A novel modification of a prompt-based approach to training encoder transformers for stance detection is proposed. It showed comparable results to those obtained with large language models, but at the cost of an order of magnitude fewer parameters. Our work tries to reliably show where do we stand in stance detection and where should we go, especially in terms of datasets and experimental settings.

ABSTRACT. We analyze 7826 publications from the International Conference on Computational Science (ICCS) between 2001 and 2023 using natural language processing and network analysis. We categorize computer science into 13 main disciplines and 102 sub-disciplines sourced from Wikipedia. After lemmatizing full texts of these papers, we calculate the similarity scores between the papers and each sub-discipline using vectors built with TF-IDF evaluation. Among the 13 main disciplines, machine learning & AI have become the most popular topics since 2019, surpassing parallel & distributed computing, which peaked in the early 2010s. Modeling & simulation, and algorithms & data structure have al-ways been popular disciplines in ICCS over the past 23 years. The most frequently researched sub-disciplines, on average, are algorithms, numerical analysis, and machine learning. Deep learning shows the most rapid growth, while parallel computing has declined over the past 23 years in ICCS publications. The network of sub-disciplines exhibits a scale-free distribution, indicating certain disciplines are more connected than others. We also present correlation analysis of sub-disciplines, both within the same main disciplines and between different main disciplines.

ABSTRACT. Protein thermostability is one of the most important features of bio-engineered proteins with significant scientific and industrial applications. Unfortunately, obtaining thermostable proteins is both expensive and complex. Artificial intelligence, or more specifically deep learning, provides novel tools that allow to predict thermostability of proteins based on their sequence. Recent advances in Protein Language Models (pLM) offer promising framework for sequence-to-sequence problems, particularly those related to melting temperature. In this work, we present EsmTemp, a transfer learning model based on the ESM-2 pLM architecture. EsmTemp undergoes training on a meticulously curated dataset comprising 24,000 protein sequences with known melting temperatures. A rigorous evaluation, conducted through a 10-fold cross-validation, yields a coefficient of determination (R2) of 0.70 and a mean absolute error of 4.3°C. These outcomes highlight how pLM has the potential to advance our understanding of protein thermostability and facilitate the rational design of enzymes for various applications.

ABSTRACT. The process of developing antenna structures typically involves prototype measurements. While accurate validation of far-field performance can be performed in dedicated facilities like anechoic chambers, high cost of construction and maintenance might not justify their use for teaching, or low-budget research scenarios. Non-anechoic experiments provide a cost-effective alternative, however the performance metrics obtained in such conditions require appropriate correction. In this paper, we consider a multitaper approach for post-processing antenna far-field characteristics measured in challenging, non-anechoic environments. The discussed algorithm enhances one-shot measurements to enable extraction of line-of-sight responses while attenuating interferences from multi-path propagation and the noise from external sources of electromagnetic radiation. The performance of the considered method has been demonstrated in uncontrolled conditions using a compact spline-based monopole. Furthermore, the approach has been favorably validated against the state-of-the-art techniques from the literature.

ABSTRACT. In this study, we delve into the realm of meta-learning to combine point base forecasts for probabilistic short-term load forecasting. Our approach encompasses the utilization of quantile linear regression, quantile regression forest, and post-processing techniques involving residual simulation to generate quantile forecasts. Furthermore, we introduce both global and local variants of meta-learning. In the local-learning mode, the meta-model is trained using patterns most similar to the query pattern. Through extensive experimental studies across 35 forecasting scenarios and employing 16 base forecasting models, our findings underscored the superiority of quantile regression forest over its competitors.

ABSTRACT. Moving toward a sustainable society requires the development of reliable indices, indicators, and computational methods that supply the tools, such as decision support systems used in assessing the achievement of sustainable development goals. The aim of this paper is to present an intelligent decision support system that enables multi-criteria evaluation, taking into account the temporal variability of the performance of the assessed alternatives. The framework of this DSS is based on the method called Data vARIability Assessment - Measurement of Alternatives and Ranking according to COmpromise Solution (DARIA-MARCOS). The proposed method was used for an exemplary multi-criteria analysis problem concerning the implementation of the sustainable development goals included in Sustainable Development Goal 11 (SDG 11), focused on sustainable cities and communities. SDG 11 aims to develop toward making cities and human settlements inclusive, safe, resilient, and sustainable. The methodical framework implemented in the demonstrated DSS ensures an efficient, automatized, and objective assessment of a multi-criteria temporal decision-making problem and gives an unequivocal, clear outcome. The results proved the usability of the developed DSS in the multi-criteria temporal evaluation of sustainable development focused on sustainable cities and communities.

ABSTRACT. Multi-Criteria Decision Analysis (MCDA) is an interdisciplinary field that addresses decision-making problems that involve multiple conflicting criteria. MCDA methods are widely applied in various domains, including medicine, management, energy, and logistics. Despite their widespread use, MCDA techniques continuously evolve to address emerging challenges.

This paper presents a new method called Stochastic Expected Solution Point SPOTIS (SESP-SPOTIS), for re-identifying MCDA models. SESP-SPOTIS conducts a stochastic search for the Expected Solution Point (ESP) which is then utilized within the Stable Preference Ordering Towards Ideal Solution (SPOTIS) framework. The study delves into comprehensive investigations of MCDA model re-identification and examines how the updated model influences the ranking of analyzed alternatives. Furthermore, the experiments were divided into training sets and tests to evaluate the similarity of the proposed approach, using rank correlation coefficients $r_w$ and $WS$. The results demonstrate that SESP-SPOTIS effectively re-identifies updated models, thereby broadening knowledge and understanding in the decision-making process of the analyzed problem. By integrating machine learning models and stochastic optimization techniques, SESP-SPOTIS contributes to advancing the methodologies for MCDA model re-identification.

ABSTRACT. Demographic information can be used to analyze processes occurring in a wide range of human social activity: in the area of management, healthcare, social secu-rity systems. One of the leading methods of demographic modeling is continuous simulation based on the system dynamics approach. Reducing computation time for fast-running continuous model implementations enables efficient development of hybrid models. In the hybrid simulation model, the discrete-event simulation approach was used as one of the components. Technical and conceptual solutions to the observed problems were presented, experiments based on demographic da-ta from the Wrocław, Poland region were performed and it was indicated that an effective hybrid simulation will allow to include additional cause-and-effect rela-tionships in the models.

ABSTRACT. Sustainability is one of the development trends of various businesses, including those focused on digital channels. One example of the practical engagement is the care taken to minimize emissions of the various greenhouse gases, e.g. carbon dioxide. While e-commerce does not directly affect emissions, it does consume electricity, the generation of which increases the amount of CO_2 in the atmosphere. Measuring the carbon footprint is one way to analyze the environmental impact of business. It also allows the comparison of different solutions and the planning of actions to minimize the negative impact of human activities. In the case of e-commerce, one method of calculating the carbon footprint is to convert the energy used to transmit information between an online store and a customer into a carbon dioxide equivalent. This paper focuses on analyzing the carbon footprint of the top 100 most popular Polish e-shops in order to verify their commitment to sustainability. The research uses a measurement method used in online carbon footprint calculators, which, despite significant simplifications, allows a rough estimate of a website's impact on carbon dioxide emissions. Nevertheless, the perceived limitations of the algorithm used made it possible to suggest directions for its development, which could significantly affect the accuracy of the calculations.

ABSTRACT. Temporal expression extraction (TEE) plays a crucial role in natural language processing (NLP) tasks, enabling the capture of temporal information for downstream tasks such as logical reasoning and information retrieval. However, current TEE research mainly focuses on resource-rich languages like English, leaving a gap for minor languages (e.g. Uyghur) in research. To address these issues, we create an English-Uyghur cross-lingual dataset specifically for the task of temporal expression extraction in Uyghur. Besides, considering the unique characteristics of Uyghur, we propose XLTU, a Cross-Lingual model in Temporal expression extraction for Uyghur, and utilize multi-task learning to help transfer the knowledge from English to Uyghur. We compare XLTU with different models on our dataset, and the results demonstrate that our model XLTU achieves the SOTA results on various evaluation metrics. We make our code and dataset publicly available.

ABSTRACT. Computational Fluid Dynamics (CFD) is a field present in a wide variety of industrial applications nowadays. Many CFD codes use the Finite Volume Method for solving the Navier-Stokes equations, which implies fine spatial discretizations for reconstructing the smallest structures of the flow. In most cases, the huge computational cost of resolving such fine discretizations using direct numerical simulation (DNS) makes the technology unaffordable. Alternatives like Large Eddy Simulation (LES) or Reynolds-Averaged Navier-Stokes (RANS) have emerged, where the smallest flow structures are modelled. However, the computational cost could be prohibitive even in this cases due to the characteristics of the problem. In recent years, Deep Learning (DL) has arisen as a promising tool for enhancing CFD simulations, both for reducing the computational cost and improving the accuracy of the solution of the traditional solvers.

In this work, we present an OpenFOAM-embedded deep learning framework for reducing the spatial discretization error in coarse meshes. We substitute the traditional differencing scheme for the convective term by a feed-forward neural network. The convective differencing scheme is an operator that relates a face's velocity with the cell's center velocities of its neighbourhood. A correct selection of this scheme is crucial for obtaining accurate and stable solutions, so numerous versions may be found in the literature depending on flow conditions. Inspired by super-resolution techniques employed in image processing, our model is trained with high-fidelity data. Thus, it generates approximations of the fine data using a coarse discretization. As a major advantage, our model enforces physical constraints of the governing equations while enriching certain parts of the traditional solver with high-resolution data. However, the main limitation is the necessity of data with an image format, that is, structured data generated from Cartesian meshes with cells of the same size.

In contrast to previous works, we embed the open-source CFD code OpenFOAM in the training loop to exploit the vast range of physics and numerical solvers this program has implemented. This allows us to extend the use of the method to non-trivial problems, becoming a general approach applicable to any fluid case simulated in OpenFOAM. This feature requires an end-to-end differentiable framework, which we obtain through the automatic differentiation of the CFD physics using a discrete adjoint version of the code. Besides, although our neural network's architecture exploits the local features of the flow to speed up the training process, we also developed a fast and bidirectional communication method between the ML framework (Python) and the CFD code (C++) to accelerate the learning process.

We test our model in a well-known industrial problem consisting of a flow passing a square cylinder. We reach reductions of the error to about 50% compared to the traditional solver in the x- and y-velocity components using an 8x coarser mesh. The training process takes an affordable amount of time and data samples. The sensitive analysis carried out shows that the error reduction is even higher in coarser meshes, as well as a faster training time. The model shows correct generalization, producing stable predictions for mid-term simulations.

ABSTRACT. We discuss Robust Variational Physics-Informed Neural Networks (RVPINNs), highlighting challenges with gradient-based optimizers like Adam and proposing a hybrid Adam/LS solver for improved convergence. The architecture involves fully-connected feed-forward neural networks with non-linear activations, emphasizing the significance of the last hidden layer's parameters. By reformulating the loss function in an ultraweak form, computational costs are reduced without compromising performance. Numerical examples in 1D and 2D demonstrate significant speed and cost improvements compared to conventional methods, showcasing the efficacy of the ultraweak implementation. Overall, RVPINNs with the hybrid solver offer faster convergence and reduced computation times, making them comparable to Adam alone.

ABSTRACT. Structural Health Monitoring (SHM) practices in full-scale operative systems such as bridges are often constrained by the lack of experimental damage scenarios, which enable detecting departures from what is considered the reference condition. If a more exhaustive assessment is desired, we need to complement the data scarcity in the experimental domain with a new source of information: computer simulations. However, validating these methods in damage detection is still a very controversial task, given the lack of real damage to fully demonstrate the capability of the proposed algorithms. Here, we explore the reliability of an unsupervised learning approach based on an Autoencoder neural network designed in our previous works. The methodology applies to a full-scale case study with available long-term monitoring data: the Infante dom Henrique bridge in Porto. The available measurements consist of postprocessed 15-minute-long acceleration signals that yield the first four vertical bending eigenfrequencies and eigenmodes. Compared to our previous work, we now train the autoencoder with experimental and synthetic healthy scenarios generated with a Finite Element (FE) model of the bridge instead of relying on experimental data only. Throughout the validation process, we evaluate the limits and confidence of the diagnostics provided by the damage detection tool. We analyze the sensitivity of the damage indicator (the averaged reconstruction error), statistically analyzing the attained performance with four testing datasets that include healthy and damaged scenarios. As a realistic representation of damage, we employ some experimental measurements and affect them with a multiplication factor obtained from specific damage simulations, which incorporate the effect of measurement error and varying environmental conditions in the damage cases. The analysis demonstrates the sensitivity of the damage indicator to synthetic data, given the limited variability of these scenarios. However, when we explore the experimental datasets, we observe that false positives and false negatives arise. Figure 1 reflects these results as separate histograms for each dataset. We also investigate the impact of damage location on the sensitivity of the input features by comparing damaged scenarios occurring at eight different deck regions. Results reveal the limitations of the unsupervised learning approach in detecting some damage scenarios, mainly for some specific deck locations. These outcomes suggest the need for complementary sensing devices at specific locations where dynamic properties (mainly the mode shape curvature) are less sensitive.

ABSTRACT. Quantum algorithms, represented as quantum circuits, can be used as benchmarks for assessing the performance of quantum systems. Existing datasets, widely utilized in the field, suffer from limitations in size and versatility, leading researchers to employ randomly generated circuits. Random circuits are, however, not representative benchmarks as they lack the inherent properties of real quantum algorithms for which the quantum systems are manufactured. This shortage of `useful' quantum benchmarks poses a challenge to advancing the development and comparison of quantum compilers and hardware.

This research aims to enhance the existing quantum circuit datasets by generating what we refer to as `realistic-looking' circuits by employing the Transformer machine learning architecture. For this purpose, we introduce KetGPT, a tool that generates synthetic circuits in OpenQASM language, whose structure is based on quantum circuits derived from existing quantum algorithms and follows the typical patterns of human-written algorithm-based code (e.g., order of gates and qubits). Our three-fold verification process, involving manual inspection and Qiskit framework execution, transformer-based classification, and structural analysis, demonstrates the efficacy of KetGPT in producing large amounts of additional circuits that closely align with algorithm-based structures. Beyond benchmarking, we envision KetGPT contributing substantially to AI-driven quantum compilers and systems.

ABSTRACT. This paper explains the main design decisions in the development of variational quantum time series models and denoising quantum time series autoencoders. Although we cover a specific type of quantum model, the problems and solutions are generally applicable to many other methods of time series analysis. The paper highlights the benefits and weaknesses of alternative approaches to designing a model, its data encoding and decoding, ansatz and its parameters, measurements and their interpretation, and quantum model optimisation. Practical issues in training and execution of quantum time series models on simulators, including those that are CPU and GPU based, as well as their deployment on quantum machines, are also explored. All experimental results are evaluated, and the final recommendations are provided for the developers of quantum models focused on time series analysis.

ABSTRACT. Application of post-quantum algorithms in newly deployed cryptosystems is necessary nowadays. In the NIST Post-Quantum Competition several algorithms that seem to be resistant against attacks mounted using quantum computers have been chosen as finalists. However, it is worth noting that one of finalists — SIKE — was catastrophically broken by a classical attack of Castryck and Decru only a month after qualifying for the final round. This shows that absolute trust cannot yet be placed in the algorithms being standardized. And so a proposition was made to use the novel, post-quantum schemes alongside the well-studied classical ones with parameters chosen appropriately to remain secure against quantum attacks at least temporarily, i.e., until a large enough quantum computer is built. This paper analyzes which classical public-key algorithms should be used in tandem with the post-quantum instances, and studies how to ensure appropriate levels of both classical and quantum security. Projections about the development of quantum computers are reviewed in the context of selecting the parameters of the classical schemes such as to provide quantum resistance for a specified amount of time.

ABSTRACT. Due to Shor's and Grover's algorithms, quantum computing has become one of the fastest-evolving areas in computational science. Nowadays, one of the most interesting branches of quantum computing is quantum annealing. This paper presents the efficient method of transforming stream cipher E0 to the QUBO problem and then retrieving the Encryption Key of this cipher using quantum annealing. Comparably to other asymmetric and symmetric cryptographic algorithms, the presented transformation is efficient, requiring only 2,728 (2,751) logical variables for attack with 128 (129) consecutive keystream bits. According to our knowledge, it is the most efficient algorithm transformation with a 128-bit key. Moreover, we show that using current quantum annealers, one can embed the attack for E0 for 58 consecutive bits of keystream, from 128 (129), which are necessary for the attack's first stage (second stage). Therefore, it is likely that it will be possible to embed E0 on available quantum annealers in the next few years.

ABSTRACT. We have developed an R-shiny, online case-based, computational, multi-methods approach to systems mapping for helping policy evaluators engage their data, including qualitative data. For non-numeric data this approach employs a Fuzzy-set algebraic approach that converts qualitative policy evaluation data into a format that can then be modelled. Be it numeric or Boolean, our approach uses a suite of tools, including machine learning, k-means cluster analysis and hierarchical cluster analysis, to produce an immediate systems map for exploration. The systems map is generated using zero-order correlations and can be explored using several network mapping tools, including in-degree/out-degree, ego-network analysis, etc. It can also be modified, or added to, to explore barriers or levers to change, and used in the standard manner of most systems mapping exercises. To demonstrate the value of our approach, we use policy data on urban planning from several European cities.

ABSTRACT. UN Security Council resolutions are central to addressing global peace and security concerns and serve as authoritative directives under the UN Charter. These resolutions authorise various actions, such as peacekeeping operations and sanctions, and carry significant normative weight, shaping international legal norms. In this study, we analyse Security Council resolutions and the evolving dynamics of multilateral diplomacy over time. Using graph analysis and natural language processing techniques, we identify clusters representing different international crises and issues, shedding light on the heterogeneity within the Council’s citation patterns and deliberative dynamics. This interdisciplinary approach provides valuable insights into the complexity of multilateralism and global governance, highlighting how the structure of deliberation changes over time and across issues.

ABSTRACT. Consensus-building can be considered a cornerstone of democratic societies and effective governance. Recently, it has been shown that social contagions can shape some forms of consensus-based collective decision-making [1]. The topology of the network underpinning such processes [2] plays a key role in promoting or hampering simple contagions—based on pairwise interactions—and complex contagions, which require social influence and reinforcement [3]. However, considering the ubiquitous scalefreeness of most social networks, it becomes imperative to delve into how this particular characteristic of networks impacts the efficacy of social contagions and the subsequent group consensus. Two specific aspects worth analyzing are: (i) understanding the interplay between scalefreeness and the transition from a simple to a complex contagion, and (ii) exploring the specificities associated with highly clustered networks. We use the leader-follower consensus model—a paradigmatic approach to collective decision-making—to systematically explore the transition from simple to complex contagions in the presence of a tunable family of synthetic scale-free networks. We uncover unique features associated with the scalefreeness of networks on the cascading dynamics of complex contagions. These findings carry profound implications for the development of innovative strategies aimed at fostering consensus within social groups.

ABSTRACT. This paper explores the detection of negativity and dissensus in international climate negotiations, focusing on the sessions of the Intergovernmental Panel on Climate Change (IPCC) from 2001 to 2022. It addresses the challenges posed by subtle diplomatic language, which hinders the straightforward application of standard sentiment analysis tools. By leveraging fine-tuned large language models, particularly a DistilBERT model trained on the Stanford Sentiment Treebank, we uncover patterns of tonal shifts and dissensus that are consistent with recent observations. climate change research. Additionally, we employ Latent Dirichlet Allocation (LDA) to model the topics discussed in IPCC sessions, identifying specific nations, like Saudi Arabia, and topics associated with negotiation obstructions. Despite the subtle language, topic modeling combined with sentiment analysis offers valuable insights into the negotiation dynamics.

ABSTRACT. In this work, we present a textual and network analysis approach to understanding the phenomenon of gridlock within the United Nations Security Council (UNSC) from 1946 to 2023. By leveraging a comprehensive dataset of resolutions and voting records, we aim to explore the dynamics and evolution of gridlock in this international body. Through the application of network analysis and machine learning techniques, we identify patterns and shifts in the council's decision-making processes.

Our study highlights how gridlock has manifested in the UNSC's operations over the decades, correlating significant political events with changes in resolution adoption rates, veto patterns, and textual complexities. We identify metrics for assessing the effectiveness of the council in overcoming gridlock, including the analysis of resolution text lengths, voting behavior, and citation networks. These metrics provide insights into the operational challenges faced by the UNSC in maintaining global peace and security.

Our results support the notion that the periods of intense geopolitical tension are associated with phases of gridlock. Notably, the here presented data-driven analysis confirms that changes in the geopolitical landscape, such as the end of the Cold War, had profound impacts on the council's functionality. Furthermore, our analysis employs textual analysis, specifically natural language processing, to perform a location analysis. This approach highlights the geographical focuses and identifies shifts in the UNSC's agenda over time, illustrating how the council's attention to specific regions has evolved.

This work contributes to the field of international relations by providing a data-driven understanding of how gridlock can affect global governance in the context of a key international organization, the UNSC. The findings may offer new tools for predicting and managing gridlock within the UNSC and other international organizations. Additionally, this approach demonstrates the potential of data science in enriching traditional qualitative analyses of international organizations.

ABSTRACT. Artificial intelligence have contributed to advancements across various industries. However, the rapid growth of artificial intelligence technologies also raises concerns about their environmental impact, due to associated carbon footprints to train computational models. Fetal brain segmentation in medical imaging is challenging due to the small size of the fetal brain and the limited image quality of fast 2D sequences. Deep neural networks are a promising method to overcome this challenge. In this context, the construction of larger models requires extensive data and computing power, leading to high energy consumption. Our study aims to explore model architectures and compression techniques that promote energy efficiency by optimizing the trade-off between accuracy and energy consumption through various strategies such as lightweight network design, architecture search, and optimized distributed training tools. We have identified several effective strategies including optimization of data loading, modern optimizers, distributed training strategy implementation, and reduced floating point operations precision usage with light model architectures while tuning parameters according to available computer resources. Our findings demonstrate that these methods lead to satisfactory model performance with low energy consumption during deep neural network training for medical image segmentation.

ABSTRACT. This study introduces an innovative memetic algorithm for optimizing the consensus of well-adapted techniques for the inference of gene regulation networks. Building on the methodology of a previous proposal (GENECI), this research adds a local search phase that incorporates prior knowledge about gene interactions, thereby enhancing the optimization process under the influence of domain expert. The algorithm focuses on the evaluation of candidate solutions through a detailed evolutionary process, where known gene interactions guide the evolution of such solutions (individuals). This approach was subjected to rigorous testing using benchmarks from editions 3 and 4 of the DREAM challenges and the yeast network of IRMA, demonstrating a significant improvement in accuracy compared to previous related approaches. The results highlight the effectiveness of the algorithm, even when only 5\% of the known interactions are used as a reference. This advancement represents a significant step in the inference of gene regulation networks, providing a more precise and adaptable tool for genomic research.

ABSTRACT. Parkinson's disease (PD) is a neurological disorder defined by the gradual loss of dopaminergic neurons in the substantia nigra pars compacta, which causes both motor and non-motor symptoms. Understanding the neuronal processes that underlie PD is critical for creating successful therapies. This work presents a novel strategy that combines machine learning (ML) and stochastic modelling with connectomic data to better understand the complicated brain pathways involved in PD pathogenesis. We use modern computational methods to study large-scale neural networks to identify neuronal activity patterns related to PD development. We aim to define the subtle structural and functional connection changes in PD brains by combining connectomic and stochastic noises. Importantly, stochastic modelling approaches reflect brain dynamics' intrinsic variability and unpredictability, shedding light on the origin and spread of pathogenic events in PD. We created a hybrid modelling formalism and a novel co-simulation approach to identify the effect of stochastic noises on the cortex-BG-thalamus (CBGTH) brain network model in a large-scale brain connectome. We use data from the Human Connectome Project (HCP) to elucidate a stochastic influence on the brain network model. Furthermore, we choose areas of the parameter space that reflect both healthy and Parkinsonian states, as well as the impact of deep brain stimulation (DBS) on the subthalamic nucleus and thalamus. We infer that thalamus activity increases with stochastic disturbances, even in the presence of DBS. We predicted that lowering the effect of stochastic noises would increase the healthy state of the brain. This work aims to unravel the complicated dynamics of neuronal activity in PD, therefore opening up new options for therapeutic intervention and tailored therapy.

ABSTRACT. The paper proposes a method of human sex recognition using individual gait features extracted by measures describing the dimensionality and uncertainty of non-linear dynamical systems. The correlation dimension and sample entropy are computed for time series representing angles of skeletal body joints as well as whole-body orientation and translation. Two aggregations strategies for pose parameters are used. In the first variant, the distinction between females and males is performed by thresholding the obtained measure values. Moreover, the supervised classification is carried out for the complex gait descriptors characterizing the movements of all bone segments. In the validation experiments, highly precise motion capture measurements containing data of 25 and 30 female and male individuals are used. The obtained, at least promising, performance assessed by correct classification rate, the area under the receiver operating characteristic curve, and average precision, higher than 89\%, 96\%, and 96\%, respectively, exceeds our expectations. Moreover, the classification accuracy based ranking of skeletal joints, as well as whole-body orientation and translation evaluating sex-discriminative traits incorporated in the movements of bone segments, is formed.

ABSTRACT. Negation detection refers to the automatic identification of linguistic expression that convey negation within a textual content. In medical and biomedical context, the negation detection plays a pivotal role in understanding clinical documentation and extracting meaningful insights. In this paper, we survey 16 articles published from 2005 to 2023 and focusing on negation detection within medical domain. Our evaluation framework encompass both methodological aspects and application-oriented considerations. Specifically, we discuss the used approaches, the emplyed methodology, the specific tasks addressed, the target language of textual analysis, and the evaluation metrics used. On the application front, for each reviewed study, we delineate the medical domains under investigation (e.g., cardiology, oncology), the types of data analyzed, and the availability of datasets. The majority of reviewed works are conducted in English, with a prevalence of machine learning and deep learning approaches, and classic classification evaluation metrics. Application domains exhibit heterogeneity, with a slight predominance in oncology, and diverse data sources including EHRs, abstracts, scientific papers, and web-derived information (e.g., Wikipedia or blog entries). Throughout this review, we will identify limitations and gaps in this research area, as well as examine the benefits it could bring to the scientific community and the methods currently employed.

ABSTRACT. Given the increasingly pervasive role and growing importance of computing and data in all aspects of science and society fundamental advances in computer science and their translation to the real world have become essential. Consequently, there may be benefits to formalizing Translational Computer Science (TCS) to complement the traditional foundational and applied modes of computer science research, as has been done for translational medicine. TCS has the potential to accelerate the impact of computer science research overall. In this talk I discuss the attributes of TCS, and formally define it. I enumerate a number of roadblocks that have limited its adoption to date and sketch a path forward. Finally, I will provide some specific examples of translational research underpinning computational science projects and illustrate the advantages to both computer science and the application domains.

ABSTRACT. Subatomic particle track reconstruction (tracking) is a vital task in High-Energy Physics experiments. Tracking is exceptionally computationally challenging and fielded solutions, relying on traditional algorithms, do not scale linearly. Machine Learning (ML) assisted solutions are a promising answer. We argue that a complexity-reduced problem description and the data representing it, will facilitate the solution exploration workflow. We provide the REDuced VIrtual Detector (REDVID) as a complexity-reduced detector model and particle collision event simulator combo. REDVID is intended as a simulation-in-the-loop, to both generate synthetic data efficiently and to simplify the challenge of ML model design. The fully parametric nature of our tool, with regards to system-level configuration, while in contrast to physics-accurate simulations, allows for the generation of simplified data for research and education, at different levels. Resulting from the reduced complexity, we showcase the computational efficiency of REDVID by providing the computational cost figures for a multitude of simulation benchmarks. As a simulation and a generative tool for ML-assisted solution design, REDVID is highly flexible, reusable and open-source. Reference data sets generated with REDVID are publicly available. Data generated using REDVID has enabled rapid development of multiple novel ML model designs, which is currently ongoing.

ABSTRACT. This paper introduces the Interpoint Inception Distance (IID) as a new approach for evaluating deep generative models. It is based on reducing the measurement discrepancy between multidimensional feature distributions to one-dimensional interpoint comparisons. Our method provides a general tool for deriving a wide range of evaluation measures. The Cramer Interpoint Inception Distance (CIID) is notable for its theoretical properties, including a Gaussian-free structure of feature distribution and a strongly consistent estimator with unbiased gradients. Our experiments, conducted on both synthetic and large-scale real or generated data, suggest that CIID is a promising competitor to the Frechet Inception Distance (FID), which is currently the primary metric for evaluating deep generative models.

ABSTRACT. Pre-trained language models encounter a bottleneck in production due to their high computational cost. Model compression methods have emerged as critical technologies for overcoming this bottleneck. As a popular compression method, knowledge distillation transfers knowledge from a large (teacher) model to a small (student) model. However, existing methods perform distillation on the entire data, which easily leads to repetitive learning for the student. Furthermore, the capacity gap between the teacher and student hinders knowledge transfer. To address these issues, we propose Data-efficient Knowledge Distillation (DeKD) with teacher assistant-based dynamic objective alignment from the data, model, and objective aspects, which empowers the student to dynamically adjust the learning process. Specifically, we first design an entropy-based strategy to select informative instances at the data level, which can reduce the learning from the mastered instances for the student. Next, we introduce the teacher assistant as an auxiliary model for the student at the model level to mitigate the degradation of distillation performance. Finally, we further develop the mechanism of dynamically aligning intermediate representations of the teacher to ensure effective knowledge transfer at the objective level. Extensive experiments on the benchmark datasets show that our method outperforms the state-of-the-art methods.

ABSTRACT. The paper presents a methodology for building concurrent models of decision-making systems based on knowledge extracted from empirical data. We assume that the data is represented by a decision table, while the decision-making system is represented by a Petri net. Decision tables contain conditional attribute values obtained from measurements or other sources. A Petri net is constructed using all true and acceptable rules generated from a given decision table. Rule factors and other parameters needed to build the net model are also computed from the data table. Three operators In, Trs and Out interpreted as uninorms are used to describe the dynamics of the net model. The expected behavior of the model is achieved by proper organization of its work. The theoretical basis of the methodology is the concepts, methods and algorithms derived from the theory of rough sets, fuzzy sets and Petri nets.

ABSTRACT. Design of ultra-wideband antennas for in-door localization applications is a challenging task that involves development of geometry that ensures appropriate balance between the size and performance. In this work, a topologically-flexible monopole has been generated using a stratified framework which embeds a gradient-based trust-region (TR) optimization algorithm in a meta-loop that gradually increases structure dimensionality. The optimization has been performed using a composite objective function that maintains acceptable size/performance trade-off. The final design features a reflection below –10 dB within the UWB spectrum and a small footprint of only 182 mm2. The considered method has been benchmarked against a standard TR-based routine executed directly on a multi-dimensional representation of the antenna model.

ABSTRACT. Physics-Informed Neural Networks (PINNs) have emerged as a powerful tool for solving partial differential equations (PDEs) in various scientific and engineering domains. However, traditional PINN architectures typically rely on fully connected multilayer perceptrons (MLPs), lacking the sparsity and modularity inherent in many numerical solvers. This study investigates a novel approach by merging established PINN methodologies with brain-inspired neural network techniques to address this architectural limitation. We use Brain-Inspired Modular Training (BIMT), leveraging concepts such as locality, sparsity, and modularity inspired by the organization of the brain. Through BIMT, we demonstrate the evolution of PINN architectures from fully connected structures to highly sparse and modular forms, resulting in reduced computational complexity and memory requirements. We showcase the efficacy of this approach by solving differential equations with varying spectral components, revealing insights into the spectral bias phenomenon and its impact on neural network architecture. Moreover, we derive basic PINN building blocks through BIMT training on simple problems akin to convolutional and attention modules in deep neural networks, enabling the construction of modular PINN architectures. Our experiments show that these modular architectures offer improved accuracy compared to traditional fully connected MLP PINNs, showcasing their potential for enhancing PINN performance while reducing computational overhead. Overall, this study contributes to advancing the understanding and development of efficient and effective neural network architectures for solving PDEs, bridging the gap between PINNs and traditional numerical methods.

ABSTRACT. It has been demonstrated that neural stochastic differential equations (neural SDE) are effective in modelling time series data with complex dynamics, particularly for those that exhibit random behaviour. In this work, we proposed a novel type of normalising flow that aims to capture the multivariate structure inherent in time series data. The gen- eral framework involves a latent process driven by a neural SDE based on the Merton jump diffusion. We implement a maximum likelihood estima- tion method for the Merton model to estimate the intensity parameter of the Poisson process in neural SDE, and achieve improved results com- pared to previous approaches in terms of generating time series data for various quality metrics. Our experiments demonstrate that the suggested approach accurately replicates both the temporal and dimensional distri- butions of the original multivariate time series. Additionally, we propose a new method for assessing the quality of synthetic time series using a Wasserstein-based similarity measure. This method is based on com- paring the distributions of signature cross-sections of the original and generated time series.

ABSTRACT. The paper investigates the stability of text-to-text T5 models in keyword generation tasks, highlighting the sensitivity of their results to subtle experimental variations such as the seed used to shuffle fine-tuning data. The authors advocate for incorporating error bars and standard deviations when reporting results to account for this variability, which is common practice in other domains of data science, but not common in keyphrase generation. Through experiments with T5 models, they demonstrate how small changes in experimental conditions can lead to significant variations in model performance, particularly with larger model sizes. Furthermore, they analyze the coherence within a family of models and propose novel approaches to assess the stability of the model. In general, the findings underscore the importance of considering experimental variability when evaluating and comparing text-to-text models for keyword generation tasks.

ABSTRACT. Dictionary-based methods are one of the strategies that have grown in the realm of time series classification. These methods involve transforming time series data into word segments, which are then applied to time series classification. Particularly, these methods are effective for time series data that have different lengths. Our contribution involves introducing the integration of dictionary-based techniques with stacked ensemble learning. This study is unique since it combines the symbolic aggregate approximation (SAX) with stacking gated recurrent units (GRU) and a convolutional neural network (CNN), referred to as SGCNN, which has not been previously investigated in time series classification. Our approach uses the SAX technique to transform unprocessed numerical data into a symbolic representation. Next, a series of words is created by segmenting the symbolic time series. The classification process is done using the SGCNN classifier. We applied it to 30 well-known benchmark datasets and conducted a comparative analysis with state-of-the-art methods. Empirical experiments demonstrate that our approach performs admirably across various datasets. In particular, our method achieves the second position among current advanced dictionary-based methods.

ABSTRACT. This is a study of numerical precision errors in various representations of 2D hyperbolic geometry. It is generally the best to combine a representation with tiling; the tests take this into account.

ABSTRACT. Turbulent flows play important roles in many flow-related phenomena that appear in various fields. However, despite numerous studies on turbulence, the nature of turbulence has not yet been fully clarified. Direct numerical simulation (DNS) of incompressible homogeneous turbulence in a periodic box is currently a powerful method for studying turbulent flows. However, even modern world-class supercomputers do not have sufficient computational resources to carry out DNS at very high Reynolds number (Re). Memory capacity constraints are particularly severe. Therefore, we have developed an out-of-core DNS (ooc-DNS) code that uses storage to overcome memory limitations. The ooc-DNS code can reduce memory usage by up to a quarter and allows DNS at a higher Re, which would be impossible under normal usage due to memory limitations. When implementing the ooc-DNS code, however, it is crucial to accelerate file input/output (I/O) because the I/O time for storage accounts for a large percentage of the execution time. In this paper, we evaluate the I/O performance of the ooc-DNS code when using a file system called the Lightweight Layered I/O Accelerator of the supercomputer Fugaku. We also evaluate the impact of the I/O cache size on I/O performance and show that the I/O speed can be accelerated by optimizing the I/O cache size. As a result, by taking on I/O cache size when executing the ooc-DNS code with $8,192^3$ grid points, the I/O speed and overall execution speed were increased by 2.4 times and 1.9 times compared to that without the I/O cache.

ABSTRACT. The aim of this paper is to show that Goertzel and Reinsch algorithms for computing trigonometric sums can be efficiently vector- ized using Intel AVX-512 intrinsics in order to utilize SIMD extensions of modern processors. Numerical experiments show that the new vectorized implementations of the algorithms using only one core achieve very good speedup over their sequential versions. The new algorithms can also be parallelized using OpenMP in order to utilize multiple cores. For sufficiently large problem sizes, the parallel implementations of the algorithms achieve reasonable speedup against the vectorized ones.

ABSTRACT. The quantum volume is a comprehensive, single-number met- ric to describe the computational power of a quantum computer. It has grown exponentially in the recent past. In this study we will assume this remains the case and translate this development into the performance development of another quantum algorithms, quantum amplitude esti- mation. This is done using a noise model that estimates the error prob- ability of a single run of an algorithm. Its parameters are related to the quantum volume under the mode’s assumptions. Applying the same noise model to quantum amplitude estimation, it is possible to relate the error rate to the generated Fisher information per second, which is the main performance metric of quantum amplitude es- timation as a numerical integration technique. This provides a prediction of its integration capabilities and shows that quantum amplitude estima- tion as a numerical integration technique will not provide an advantage over classical alternatives without major breakthroughs.

ABSTRACT. While the majority of focus in quantum computing has so far been on monolithic quantum systems, quantum communication networks and the quantum internet in particular are increasingly receiving attention from researchers and industry alike. The quantum internet may allow a plethora of applications such as distributed or blind quantum computing, though research still is at an early stage, both for its physical implementation as well as algorithms; thus suitable applications are an open research question. We evaluate a potential application for the quantum internet, namely quantum federated learning. We run experiments under different settings in various scenarios (e.g. network constraints) using several datasets from different domains and show that (1) quantum federated learning is a valid alternative for regular training and (2) network topology and nature of training are crucial considerations as they may drastically influence the models performance. The results indicate that more comprehensive research is required to optimally deploy quantum federated learning on a potential quantum internet.

ABSTRACT. Connecting quantum computers to a quantum network opens a wide array of new applications, such as securely performing computations on distributed data sets. Nearterm quantum networks are noisy, however, and hence correctness and security of protocols are not guaranteed. To study the impact of noise, we consider a multiparty summation protocol with imperfect shared entangled states. We study analytically the impact of both depolarising and dephasing noise on this protocol and the noise patterns arising in the probability distributions. We conclude by eliminating the need for a trusted third party in the protocol using Shamir’s secret sharing.

ABSTRACT. Optimizing quantum circuits and reducing errors plays a crucial role in quantum circuit computation. One of the efficient representations of quantum circuits is by algebraic expressions. Our proposed method directly de-rives algebraic expressions, ensuring firstly, that parallelism is maximized, that is, the number of slices is minimized, and secondly, the computation required for obtaining the desired algebraic expression is reduced. This results in quantum circuits that are more efficient in term of space and in terms of computation time. Once we acknowledge that every quantum circuit can be represented using algebraic expressions, our objective is to ex-press them efficiently. This means creating a circuit, that is more compact. The size of a circuit, often measured by both the number of gates used and its depth, is a crucial factor. The simplification of algebraic expressions offers methods to streamline optimized circuits, ultimately reducing the number of gates and depth. This reduction is aimed at minimizing the overall complexity of the expressions, resulting in more efficient quantum computations. In this paper, we also show that optimized circuit will have less errors when compared to original circuits, Simulations are given.

ABSTRACT. The paper introduces a novel General Quantum Image Representation Model aimed at providing a general description for quantum image encoding and manipulation methods. It traces the evolution of quantum image representation from its inception, highlighting key methodologies and their contributions to the field.

Alongside the mathematical description authors present the programmatic implementation of the model that enables further experimentation and simplifies the development of quantum image encoding methods. The implementation, facilitated through an open-source framework available via a GitHub repository, leverages the Qiskit Python library, making it accessible for further research and development in quantum image processing. This dual approach not only provides a solid theoretical foundation but also practical tools for advancing quantum computing's applications in image processing.

ABSTRACT. Given the fragile, stochastic and time critical nature of quantum communication systems, it is useful to analyse them with the rigour of formal methods. However, computationally expensive methods like exact probabilistic model checking do not scale with the size of the quantum network. In this work, we analyse entanglement swapping, an important primitive in quantum networks, using statistical model checking. We investigate the robustness of entanglement swapping against important parameters like longevity of quantum memory, success probability of entanglement generation and Bell State Measurements, and heterogeneity of the quantum network nodes. We demonstrate the usefulness of the approach using the MultiVeStA statistical model checker and the SeQUeNCe quantum network simulator.

ABSTRACT. In the realm of energy-efficient building materials, Vacuum Insulation Panels (VIPs) have emerged as a forefront solution. Expanded perlite (EP) stands out for its unique combination of low density, cost-effectiveness, and excellent thermal insulating properties among the myriad of materials employed in VIPs. As the demand for sustainable and energy-efficient building practices continues to rise, understanding and enhancing the thermal performance of EP VIPs throughout their lifetime becomes paramount.

This study presents an integrated model utilizing analytical methods and finite element analysis (FEA) to simulate the heat transfer and predict the thermal conductivity of EP powder VIPs across varying gas pressures. It introduces a procedure to generate representative elementary areas (REAs) adaptable to various material characteristics; in comparing the simulation results to measurement values, the proposed model demonstrates reliable predictive performance from 0.0001 to 1 atm. The proposed model efficiently handles rapid thermal conductivity changes near atmospheric pressure, resolving distortion issues in other works.

Based on the model results of REAs reflecting various material characteristics, we found that reducing the non-flake ratio of particles and decreasing the thickness of flake particles obstruct the heat transfer across all pressure ranges. When the thermal conductivity of the absolute solid is relatively high, it is advisable for the industry to prioritize applying finer grinding; conversely, efforts should be directed towards reducing the thickness of flake particles. Finally, the impact of characteristic pore diameter on heat transfer should be discussed in conjunction with the pressure conditions of the material.

ABSTRACT. Bangladesh stands as a prime example of a nation exceptionally vulnerable to the adverse effects of climate change. Its low-lying coastal and deltaic landscape predisposes it to frequent flooding, a challenge exacerbated by a significant portion of its population grappling with poverty. The country is already experiencing the impacts of climate change, including more frequent and severe flooding that has led to the displacement of millions of people and has intensified existing social and economic challenges. Despite these formidable challenges, Bangladesh has also emerged as a global leader in climate resilience and preparedness, having made significant progress in reducing cyclone-related deaths and protecting its population from the consequences of climate change. Notably, non-governmental organisations, like our partners Save the Children, are keen to explore how they can support the most vulnerable communities by establishing the efficacy of current coping strategies for sustained resilience against climate change. To facilitate this, we are in the process of creating an agent-based model that examines the coping mechanisms adopted in response to climate-induced flooding in Bangladesh. This paper presents the initial phase of developing a multiscale conceptual model tailored to understanding this complex situation.

ABSTRACT. Forced migration is a displacement of people forced to flee due to armed conflict, violence, natural disaster or famine. In 2023, the United Nations Refugee Agency reported that there are more than 108.4 million people forcibly displaced worldwide including 62.5 million internally displaced persons and 35.3 million refugees [1]. Researchers have explored why forced migration occurs and what effects it has on economies. However, predicting forced displacement movements is important today, as it can help governments and NGOs to effectively assist vulnerable migrants and efficiently allocate humanitarian resources.

Suleimenova et al. [2] proposed an agent-based simulation toolkit, namely Flee, to predict arrivals of forcibly displaced people in neighbouring countries and analysed the sensitivity of input parameters to identify assumptions that are particularly pivotal to the validation results [3]. Subsequently, we improved and expanded Flee’s applicability to various conflict situation by introducing adaptable rules for agent movement and creation, along with factors such as ethnicity, religion, gender, and/or age. Hence, we aim to investigate the sensitivity of these new input parameters using the FabFlee plugin of FabSim3 [4], which is an automation tool, and EasyVVUQ [5] to perform Sobol’s method [6] for sensitivity analysis. We analyse key parameters across four conflict situations of Mali, Burundi, South Sudan and Central African Republic, using stochastic collocation and polynomial chaos expansion to sample our simulation input parameters. This allows us to find highly sensitive parameters in our simulation results.

ABSTRACT. In this paper, a multi-scale boundary element method is presented for modelling damage within polycrystalline materials. The constitutive behaviour of the macro-continuum is described through micromechanics simulations utilizing averaging theorems. To prevent the localization of micro-damage at the macro-scale, an integral non-local approach is employed. At the micro-scale, the model considers multiple intergranular crack initiation and propagation under mixed mode failure conditions. Furthermore, a non-linear frictional contact analysis is utilized to model cohesive-frictional grain boundary interfaces. Both micro- and macro-scales are addressed using the boundary element method, and a scheme for coupling micro-BEM with macro-FEM is proposed. Mesh independency is investigated to demonstrate the accuracy of the method, and comparisons with macro-FEM models validate different modelling approaches. Additionally, the study explores the microstructural variability of the macro-continuum to examine potential applications in heterogeneous materials. The paper presents a three-dimensional micro-mechanical crystal-level model for analysing intergranular degradation and failure in polycrystalline materials. Using Voronoi tessellations to generate polycrystalline microstructures, the model retains key statistical features of polycrystalline aggregates. The formulation relies on a grain-boundary integral representation of the elastic problem for the aggregate crystals, which are modelled as three-dimensional anisotropic elastic domains with random orientation in space. The derivation and application of a traction-separation law to model damage initiation and evolution at grain boundaries are discussed, along with a non-linear frictional contact analysis for addressing micro-crack surfaces' behaviour. The formulation's incremental-iterative algorithm for tracking degradation and micro-crack evolution is presented and validated through numerical tests on various polycrystalline microstructures. These tests demonstrate the model's capability to track damage nucleation, evolution, and coalescence under different loading conditions.

ABSTRACT. I will discuss the quantification of uncertainty in predictive models arising in physics-based models and models based on machine-learning. Applications will include predictions of the impact of pandemics, the design of advanced materials, discovery of new drugs and the behaviour of turbulent fluids. The curse of dimensionality has hitherto circumscribed the systematic study of more complex natural and artificial systems but the advent of scalable approaches is now starting to change things. A paradigm case which is widely used within the scientific community across all fields from physics and chemistry to materials, life and medical sciences is classical molecular dynamics. I will describe how we are now able to make global rankings of the sensitivity of quantities of interest to the many hundreds to thousands of parameters which are used in these models. Virtually all approaches to uncertainty quantification make use of large scale ensemble simulations and these typically require access to very sizeable supercomputers, among which is Frontier, on which we are currently planning some major UQ campaigns in collaboration with colleagues at OLCF.

ABSTRACT. Lipid microdomains are essential structural components of cellular membranes, known for their role in the organization of membrane proteins and signaling pathways. Their identification and detailed understanding are essential to unravel the complexities of cellular functions, such as signal transduction, membrane trafficking, and the pathogenesis of various diseases. This study focuses on the organization of lipid membranes, primarily aiming to enhance our understanding of the recognition and spatial arrangement of lipid microdomains. Using a combination of molecular dynamics simulations and a suite of machine learning (ML) techniques, we analyzed the distribution and behavior of lipids in various membrane configurations. Our investigation was mainly focused on two lipid systems: POPC/PSM/CHOL, to investigate raft/nonraft interactions, and DPPC/DLIPC/CHOL, to explore the nuances between liquid ordered and liquid disordered phases. The presence of cholesterol has emerged as a vital factor in the formation of liquid order domains that influence the structural and dynamic properties. By applying ML algorithms, we achieved the accurate classification of membrane regions into liquid order, liquid-disordered, or interfacial states, demonstrating the potential of computational methods to predict complex biological organizations. The findings underscore the importance of lipid arrangements in cellular processes, offering insight into membrane behavior that could aid future therapeutic strategies, particularly for diseases affected by membrane recomposition.

ABSTRACT. Machine learning (ML) methods are powerful tools for identifying new biomarkers from high-throughput data. However, the number of relevant biomarkers extracted by these methods for further biological analysis is generally still high. Therefore, the development of more complex biomarker selection protocols based on the ML approach, with additional processing of information from biological databases (DB), is very important for the further development of molecular diagnostics and therapy.

In this study, we present EnsembleFS user-friendly R toolkit (R package and Shiny web application) for heterogeneous ensemble feature selection of molecular data that also supports users in the analysis and interpretation of the most relevant biomarkers. EnsembleFS is based on five different feature filters (U-test, minimum redundancy maximum relevance (MRMR), Monte Carlo feature selection, and multidimensional feature selection in 1D and 2D versions). It uses supervised ML methods to evaluate the quality of the set of selected features, and retrieves the biological characteristics of biomarkers online from the nine open biological DB. The functional modules to identify potential biomarker candidates, evaluation, comparison, analysis, and visualization of the model results make EnsembleFS an excellent tool for selection, random forest binary classification, and comprehensive analysis of biomarkers.

EnsembleFS R Shiny application (web demo server) is available at https://uco.uwb.edu.pl/apps/EnsembleFS, the source code at (https://github.com/biocsuwb/EnsembleFS). EnsembleFS R package is available at https://github.com/biocsuwb/EnsembleFS-package.

ABSTRACT. The analysis of Disease-Disease Associations (DDAs) and Gene-Disease Associations (GDAs) is a relevant task in bioinformatics. These are analysed to investigate the interactions between sets of diseases and genes as well as their similarity, e.g., to improve the phases of diagnosis, prognosis and treatment in medicine. Generally, the extraction of information of interest from large-scale data, usually heterogeneous and unstructured, is performed via time-consuming processes. Therefore, several computational approaches have been focused on their prediction through data integration and machine learning techniques. This paper presents a solution for Inferring DDA (IDDA) by integrating curated biomedical ontologies and medical dictionaries. It is able to extract a set of DDA using an in-house score based on the GDA. A preliminary step based on data enrichment retrieves the information about gene and disease, and it integrates these with a set of curated biological data ontologies and dictionaries. Specifically, IDDA extracts DDAs based on an in-house score, which uses GDAs for its evaluations. In a preliminary step, it performs data enrichment to retrieve concepts both for diseases and genes, by integrating several curated biomedical ontologies and medical dictionaries.

ABSTRACT. Cellular automata (CA) exemplify systems where simple local interaction rules can lead to intricate and complex emergent phenomena at large scales. The various types of dynamical behavior of CA is usually categorized empirically into Wolfram's complexity classes. Here, we propose a quantitative measure, rooted in quantum information theory, to categorize the complexity of classical deterministic cellular automata. Specifically, we construct a Matrix Product Operator (MPO) of the transition matrix on the space of all possible CA configurations. We find that the growth of entropy of the singular value spectrum of the MPO reveals the complexity of the CA and can be used to characterize its dynamical behavior. This measure defines the concept of operator entanglement for CA, demonstrating that quantum information measures can be meaningfully applied to classical deterministic systems.

ABSTRACT. In this paper we discuss computer simulations of de Sitter and anti de Sitter spacetimes, which are maximally symmetric, relativistic analogs of non-Euclidean geometries. We present prototype games played in these spacetimes, discuss the technical challenges in creating such a simulation, and discuss the geometric and relativistic effects that can be witnessed by the players of our games and visualizations.

ABSTRACT. Satellite computing system (SCS), with its huge economic value, is suffering from increasing attacks. Moving Target Defense (MTD) can create the asymmetric situation between attacks and defenses by changing the attack surface. As SCS’s limited defense resources, current MTD defense timing selection methods are not suitable for SCS. This paper proposes a Markov Game based Defense Timing Selection (MGDTS) approach for MTD in SCS. MGDTS formulates attack-defense adversarial relationship as a Markov game with incomplete information, and explicit costs are used to define the resource consumption of a defender. For defense timing decision, MGDTS uses a Markov decision process to construct the defense timing decision equation, and a real-time dynamic programming to solve the equation. Experimental results show that compared with other MTDs, MGDTS can improve the security of MTD while reducing its costs.

ABSTRACT. The development of Cloud-Fog-Edge computing infrastructures in response to the rapid advance of IoT technologies requires applications to be positioned closer to users at the edge of the network. Characterised by a geographically distributed configuration with numerous heterogeneous nodes, these infrastructures face challenges such as node failures, mobility constraints, resource limitations and network congestion. To address these issues, the adoption of microservices-based application architectures has been encouraged. However, the interdependencies and function calls between services require careful optimisation, as each has unique resource requirements. In this paper, we propose a new model and heuristic for the placement of microservices in the Cloud-Fog-Edge continuum, based on community detection and a greedy algorithm to optimise energy use while taking into account the resource constraints of the nodes and ensuring that response time is acceptable. This method aims to reduce energy consumption and network load, thereby improving the efficiency and sustainability of the IT infrastructure. Results have been compared with different scenarios and show that our approach can significantly reduce energy consumption and make efficient use of resources on several parameters.

ABSTRACT. In this extended abstract, we describe the features included in a scalable agent-based simulation of EVs that is currently under active development by our team. Its proposed applications are in 1) charging infrastructure planning at different scales (from individual buildings to districts and up to full city-scale) and 2) scenario-based analyses of the impact of high levels of electrification on the distribution grid infrastructure. We aim to demonstrate the capabilities of this tool in several case studies such as the evaluation of charger adequacy for a city with high levels of EV adoption, as well as in the optimisation of energy patterns of an electrified vehicle fleet in an airport environment.

ABSTRACT. This is a follow-up study on Zernike moments applicable in detection tasks owing to a construction of complex-valued integral images that we have proposed in [3]. The main goal of the proposition was to calculate the mentioned features fast (in constant-time). The proposed solution can be applied with success when dealing with single images, however it is still too slow to be used in real-time applications, for example in video processing. In this paper we propose a technique in order to reduce the detection time in real-time applications. The degree of reduction is controlled by two parameters: fs (related to the gap between frames that undergo a full scan) and nb (related to the size of neighborhood to be searched on non-fully scanned frames). We present a series of experiments to show how our solution performs in terms of both detection time and accuracy.

ABSTRACT. This paper presents an inference algorithm for knowledge bases with a rule cluster structure. The research includes the study of the efficiency of inference, measured by the number of cases in which the inference was successful. Finding a rule whose premises are true and activating it leads to extracting new knowledge and adding it as a fact to the knowledge base. We aim to check which clustering and inference parameters influence the inference efficiency. We used four various real datasets in our experimental stage. Overall, we proceeded with almost twenty thousand experiments. The results prove that the clustering algorithm, the amount of input data, the method of cluster representation, and the subject of clustering significantly impact the inference efficiency.

ABSTRACT. We propose a general algorithm that treats cascade training as a tree search process working according to Dijkstra's algorithm in contrast to our previous solution based on branch-and-bound technique. The reason behind the algorithm change is reduction of training time. This change does not affect in anyway quality of the final classifier. We conduct experiments on cascades trained to become face or letter detectors with Haar-like features or Zernike moments being the input information, respectively. We experiment with different tree sizes and different branching factors. Results confirm that training times of obtained cascades, especially for large heavily branched trees, were reduced. For small trees, the previous technique can sometimes achieved better results but the difference is negligible in most cases.

ABSTRACT. This contribution describes an efficient and simple preprocessing algorithm for finding a convex hull, maximum distance of points or convex hull diameter, and the smallest enclosing circle in $E^2$. The proposed algorithm is convenient for large data sets with unknown intervals and ranges of the data sets. It is based on efficient preprocessing, which significantly reduces points used in final processing by standard available algorithms.

ABSTRACT. The resiliency of criminal networks against law enforcement interventions has driven researchers to investigate methods of creating accurate simulated criminal networks. Despite these efforts, insights reaching law enforcement agencies remain general and insufficient, warranting a new approach. Therefore, we created CrimeSeen - an interactive visualization and simulation environment for exploring criminal network dynamics using computational models. CrimeSeen empowers law enforcement agencies with the possibility to independently test specific scenarios and identify the most effective disruption strategy before deploying it. CrimeSeen comprises of three components: Citadel, a web-based network visualization and simulation tool serving as the interface; the model, defining rules for criminal network dynamics over time, with the Criminal Cocaine Replacement Model as the use-case in this project; and the simulator, connecting the model and interface and enhancing their functionality through transformations, triggers, and statistics. CrimeSeen was evaluated with sequential usability testing, revealing a positive trend in effectiveness and efficiency over time, with mean scores exceeding 80%. However, user satisfaction did not significantly change and remained below the average for web applications, prompting recommendations for future work.

ABSTRACT. The article presents a modified CORDIC algorithm for implementing a Givens rotator. The CORDIC algorithm is an iterative method for computing trigonometric functions and rotating vectors without using complex calculations. The authors propose two modifications for improving the classical CORDIC algorithm: completing iterations with one-directional rotation of the vector at the final stages and choosing a scaling factor value that can be implemented with low-cost dedicated hardware utilising canonical signed digits representation. The modified algorithm is implemented in a pipeline approach using Verilog language in an Altera Cyclone V System-on-Chip FPGA. The results show that the proposed algorithm achieves higher accuracy and lower latency than the classic CORDIC algorithm.

ABSTRACT. In this paper we present a numerical feed-forward strategy for the Augmented Kalman Filter and show its application to a diffusive dominated inverse problem: heat source reconstruction from boundary measurements. The method is applicable in general to forcing terms estimation in lumped and distributed parameters models.

ABSTRACT. Single-layer shallow water models have been widely used for simulating shallow water waves over both fixed and movable beds. However, these models can not capture some hydraulic features such as small eddy currents and flow recirculations. This study presents a novel numerical approach for coupling multi-layer shallow water models with elastic deformations to accurately capture complex recirculation patterns in wind-driven flows. This class of multi-layer equations avoids the computationally demanding methods needed to solve the three-dimensional Navier-Stokes equations for free-surface flows, but it provides stratified flow velocities since the pressure distribution is still assumed to be hydrostatic. In the current study, the free-surface flow problem is approximated as a layered system made of multiple shallow water equations of different water heights but coupled through mass-exchange terms between the embedded layers. Deformations in the topography are accounted for using linear elastostatic systems for which an internal force is applied. Transfer conditions at the interface between the water surface and the topography are also developed using frictional forces and hydrostatic pressures. For the computational solver, we implement a fast and accurate hybrid finite element/finite volume method solving the linear deformations on unstructured meshes and the nonlinear flows using well-balanced discretisations. Numerical results are presented for various problems and the computed solutions demonstrate the ability of the proposed model in accurately resolving wind-driven flows over deformable topography.

ABSTRACT. CFD has emerged as an indispensable tool for comprehending and refining fluid flow phenomena within engineering domains. The recent integration of CFD with AI has unveiled novel avenues for expedited simulations and computing precision. This research paper delves into the accuracy of amalgamating CFD with AI and assesses its performance across modern server-class Intel CPU/GPU architectures such as the 4th generation of Intel Xeon Scalable CPUs (codename Sapphire Rapids) and Intel Data Center Max GPUs (or Ponte Vecchio). Our investigation focuses on exploring the potential of mixed-precision techniques with diverse number formats, namely, FP32, FP16, and BF16, to accelerate CFD computations through AI-based methods. Particular emphasis is given to validating outcomes to ensure their applicability across a CFD motorBike simulation. This research explores the performance/accuracy trade-off for both AI training and simulations, including OpenFOAM solver and interference with the trained model, across various data types available on Intel CPUs/GPUs. We aim to provide a thorough understanding of how different number formats impact the performance and accuracy of the DNNbased model in various application scenarios running on modern HPC architectures.

ABSTRACT. The Maximum Boolean Satisfiability Problem (also known as the Max-SAT problem) is the problem of determining the maximum number of disjunctive clauses that can be satisfied (i.e., made true) by an assignment of truth values to the formula’s variables. This is a generalization of the well-known Boolean Satisfiability Problem (also known as the SAT problem), the first problem that was proven to be NP-complete. With the proliferation of quantum computing, a current approach to tackle this optimization problem is Quantum Annealing (QA). In this work, we compare several gadgets that translate 3-SAT problems into Quadratic Unconstrained Binary Optimization (QUBO) problems to be able to solve them in a quantum annealer. We show the performance superiority of the not-yet-considered gadgets in comparison to state-of-the-art approaches when solving random instances in D’Wave’s quantum annealer.

ABSTRACT. Quantum annealers (QA), such as D-Wave systems, become increasingly efficient and competitive at solving combinatorial optimization problems. However, solving problems that do not directly map the chip topology remains challenging for this type of quantum computer. The creation of logical qubits as sets of interconnected physical qubits overcomes limitations imposed by the sparsity of the chip at the expense of increasing the problem size and adding new parameters to optimize. This paper explores the advantages and drawbacks provided by the structure of the logical qubits and the impact of the rescaling of coupler strength on the minimum spectral gap of Ising models. We show that densely connected logical qubits require a lower chain strength to maintain the ferromagnetic coupling. We also analyze the optimal chain strength variations considering different minor embeddings of the same instance. This experimental study suggests that the chain strength can be optimized for each instance. We design a heuristic that optimizes the chain strength using a very low number of shots during the pre-processing step. This heuristic outperforms the default method used to initialize the chain strength on D-Wave systems, increasing the quality of the best solution by up to $17.2\%$ for tested instances on the max-cut problem.

ABSTRACT. Tactical deconfliction problem involves resolving conflicts between aircraft to ensure safety while maintaining efficient trajectories. Several techniques exist to safely adjust aircraft parameters such as speed, heading angle, or flight level, with many relying on mixed-integer linear or nonlinear programming. These techniques, however, often encounter challenges in real-world applications due to computational complexity and scalability issues. This paper proposes a new quantum approach that applies the Quantum Approximate Optimization Algorithm (QAOA) and the Quantum Alternating Operator Ansatz (QAOAnsatz) to address the aircraft deconfliction problem. We present a formula for designing quantum Hamiltonians capable of handling a broad range of discretized maneuvers, with the aim of minimizing changes to original flight schedules while safely resolving conflicts. Our experiments show that a higher number of aircraft poses fewer challenges than a larger number of maneuvers. Additionally, we benchmark the newest IBM quantum processor and show that it successfully solves four out of five instances considered. Finally, we demonstrate that incorporating hard constraints into the mixer Hamiltonian makes QAOAnsatz superior to QAOA. These findings suggest quantum algorithms could be a valuable algorithmic candidate for addressing complex optimization problems in various domains, with implications for enhancing operational efficiency and safety in aviation and other sectors.

ABSTRACT. Large eddy simulations (LES) are extensively employed in aerodynamics-related industrial research and applications. However, the use of lattice-Boltzmann-based LES in hemodynamics is less widely documented. This study investigates the feasibility of employing lattice-Boltzmann-based large eddy simulation techniques, specifically the Smagorinsky-based subgrid scale turbulence model, for simulating high Reynolds number blood flow at a coarse-grained resolution. Initially, a stenotic channel flow simulation is conducted, with results undergoing validation against existing experimental data and direct numerical simulation results, showing strong agreement for both. Subsequently, our model is applied to simulate aortic stenosis at a resolution of $100 \mu m$, demonstrating the capability to model high Reynolds numbers around 4500, despite such flows conventionally requiring a resolution of around $20 \mu m$. These findings highlight the significant potential of employing LES techniques in blood flow simulations using the lattice Boltzmann method, offering advantages for large-scale human simulations at coarser resolutions.

ABSTRACT. In the modern, fast-moving, and technology-centric business world, the importance of service desks in swiftly and effectively addressing tech-related challenges cannot be overstated. The changing nature of work environments, particularly with the rise of remote and hybrid models post-pandemic, highlights the critical need for a simulation strategy that goes beyond individual departments to include the broader organizational context across various levels. Focusing on this need, we embarked on a project using Discrete Event Simulation (DES) to tackle the operational hurdles faced by the service desk of a leading UK-based telecommunications firm, marking the beginning of a larger initiative aimed at multiscale modeling and simulation (MMS). We have formulated a robust five-phase strategic framework to elevate and enhance organizational performance by applying a detailed DES analysis to refine the service desk operations. This framework examines the potential for targeted improvements within the service desk to have wider benefits, impacting everything from staff satisfaction and efficiency to the overall resilience of the organization.

ABSTRACT. The surface composition of proteins is intricate, showcasing a mix of positive and negative charges, hydrogen bonding sites, and nonpolar areas. This complexity allows proteins to interact with various molecules and surfaces through a myriad of mechanisms. Given the vast differences in size, shape, and flexibility among proteins, they cannot be treated as a homogeneous group. Therefore, generalizations about protein behavior at interfaces are not feasible. Nonetheless, there is a pressing need to develop predictive models for these systems [1]. Molecular simulation stands out as one of the most direct methods for investigating interactions of proteins with materials at the molecular level. In our studies, we concentrate on two crucial subjects concerning the interaction between proteins and the surfaces of materials: i) the interaction of serum proteins with PVC polymeric materials; and ii) the interaction of Cutinase enzyme with poly(ethylene terephthalate) (PET). These investigations were conducted using all-atom Molecular Dynamics (MD) simulations, and the significance of each case will be thoroughly discussed.

i- Adsorption of Plasma Proteins on PVC polymeric materials Understanding the physio-chemical factors driving biofouling is essential for advancing the design of biomaterials [2,3]. This study assessed the affinity of plasma proteins for Polyvinyl chloride (PVC) through molecular docking, identifying the pivotal plasma proteins for further investigation. Utilizing MD simulations, we quantitatively examined the interactions between plasma proteins and PVC, scrutinizing potential structural changes in the protein during adsorption. HSA demonstrated spontaneous adsorption on the PVC surface without significant damage to its secondary structure. Thermodynamic properties governing the adsorption process were evaluated by calculating the Potential of Mean Force (PMF) along the direction normal to the surface. Additionally, investigations at different temperatures (290 K and 310 K) consistently revealed an enthalpy-driven adsorption process. This molecular-level study offers a comprehensive evaluation of a critical process influencing medical device compatibility.

ii- Probing Enzymatic PET Degradation: Molecular Dynamics Analysis of Cutinase Adsorption and Stability Understanding the mechanisms influencing poly(ethylene terephthalate) (PET) biodegradation is crucial for developing innovative strategies to accelerate the breakdown of this persistent plastic [4,5]. In this study, we employed all-atom MD simulation to investigate the adsorption process of the LCC-ICCG cutinase enzyme onto the PET surface. Our results revealed the primary driving forces for the adsorption of the cutinase enzyme onto PET. Additionally, we studied the changes in the enzyme's tertiary and secondary structures during the interaction with PET. The significance of this study lies in unraveling the molecular intricacies of the adsorption process, providing valuable insights into the initial steps of enzymatic PET degradation.

References: [1] Charles A Haynes and Willem Norde. Globular proteins at solid/liquid interfaces. Colloids and surfaces B: Biointerfaces, 2(6):517–566, 1994. [2] D. F. Williams, Biocompatibility pathways and mechanisms for bioactive materials: The bioactivity zone, Bioact. Mater. 10:306–322, 2022. [3] J. L. Brash, T. A. Horbett, R. A. Latour, P. Tengvall, The blood compatibility challenge. part 2: Protein adsorption phenomena governing blood reactivity, Acta Biomater. 94:11–24, 2019. [4] V. ournier, C. Topham, A. Gilles, B. David, C. Folgoas, E. Moya-Leclair, E. Kamionka, M.L. Desrousseaux, H. Texier, S. Gavalda, M. Cot, E. Gu ́emard, M. Dalibey, J. Nomme, G. Cioci, S. Barbe, M. Chateau, I. Andr ́e, S. Duquesne, A. Marty, An engineered PET depolymerase to break down and recycle plastic bottles. Nature, 580:216–219, 2020. [5] P. Fayon, J. Dev ́emy, C. Emeriau-Viard, K. Ballerat-Busserolles, F. Goujon, A. Dequidt, A. Marty, P. Hauret, P. Malfreyt, Energetic and Structural Characterizations of the PET-Water Interface as a Key-step in Understanding its Depolymerization. J. Phys. Chem. B 127(15):3543-3555, 2023.

ABSTRACT. With the advent of exascale computing, the concept of digital twins is gaining significant traction in many fields of engineering and science. Within the biomedical context, human digital twins seek to drive forward personalised healthcare through the creation of digital copies of a particular individual in order to allow the optimisation of their treatment and maintenance of good health. The construction of such digital twins will require the coupling of multiple specialist simulation codes to recreate the multiscale and multicomponent complexity of the human body. Achieving this will require a platform that allows models to be ‘plugged in’ to a full human model. Ideally, such constructions would be able to be executed with minimal modification of the source codes being linked in order to maintain the modularity of the full human model and reduce maintenance requirements for the codes. In this talk, we discuss the steps made in using the MUSCLE3 platform to achieve a modular coupling with the vascular blood flow simulation code HemeLB. Some of our previous work with HemeLB has developed a self-coupled version of the code to conduct simulations of simultaneous arterial and vascular flow. This coupling used MPI communications between two HemeLB instances and we have replaced these with the MUSCLE3 framework. We have a particular interest in examining how the coupling approach used impacts on the scaling performance of HemeLB. To compare the implementations, we benchmarked the simulation performance of the two implementations using the ARCHER2 (EPCC, Edinburgh, UK) and SuperMUC-NG (LRZ, Munich, Germany) computers using a representative vascular geometry. Within this presentation, we will discuss these results and their implications for the construction of coupled models particularly with regards to the portability and ease of maintenance of the implementation in our context. Such observations will be useful to those seeking to integrate their models with a human digital twin at all scales of computation.

ABSTRACT. We have developed a novel neural network model that simulates the function of the central nervous system (CNS) in controlling neural motor sensors. This model employs reinforcement learning to mimic the process of human walking while avoiding obstacles. Through an analysis of the learning process, we have discovered that our model has the potential to replicate predicted behaviors in both healthy and impaired individuals. Notably, we have been able to reproduce behaviors associated with CNS neuroplasticity, whereby increased knowledge leads to improved learning outcomes. In contrast, our simulations of impaired individuals have reflected the reality that these individuals tend to require more repetitions to perform a given task correctly than their healthy counterparts.

ABSTRACT. Sea ice plays an essential role in global ocean circulation and in regulating Earth's climate and weather, and melt ponds that form on the ice have a profound impact on the Arctic's climate by altering the ice albedo. Melt pond evolution is complex, sub grid scale and poorly understood - and melt ponds are represented in sea ice models as parametrisations. Parametrisations of these physical processes are based on a number of assumptions and can include many uncertain parameters that have a substantial effect on the simulated evolution of the melt ponds. 

We have shown, using Sobol sensitivity analysis and through investigating perturbed parameter ensembles (PPEs), that a state-of-the-art sea ice column model, Icepack, demonstrates substantial sensitivity to its uncertain melt pond parameters. These PPEs demonstrate that perturbing melt pond parameters (within known ranges of uncertainty) cause predicted sea ice thickness over the Arctic Ocean to differ by many metres after only a decade of simulation. Understanding the sources of uncertainty, improving parametrisations and fine tuning the parameters is a paramount, but usually very complex and difficult task. Given this uncertainty, we propose to replace the sub grid scale melt pond parametrisation (MPP) in Icepack with a machine learning emulator. 

Building and replacing the MPP with a machine learning emulator has been done in two broad steps that contain multiple computational challenges. The first is generating a melt pond emulator using 'perfect' or 'model' data. Here we demonstrate a proof of concept and show how we achieve numerically stable simulations of Icepack when embedding an emulator in place of the MPP - with Icepack running stably for the whole length of the simulations (over a decade) across the Arctic. 

Secondly, we develop and discuss an offline an emulator built from observational data that faithfully predicts the melt pond state given climatological input variables. Embedding an observational emulator can require different challenges as compared to using model data, such as not all variables needed by the host model being observable/observed for an emulator to predict. We discuss the computational challenges to be faced in interfacing this emulator with a sea ice model. Our work contributes to a broader discussion on how data driven neural networks can replace or compliment the `parametric’ sub grid scale parametrisation approach in modelling of dynamical systems.

ABSTRACT. Data assimilation (DA) is a Bayesian method to combine imperfect observations and a model generated prior to obtain an improved estimate for the true state of the system. It is envisaged that neXtSIM_DG, a new sea ice model, is going to be endowed with a DA system. Most of the current DA approaches assume that errors follow a Gaussian distribution. This assumption cannot hold in sea ice models because some sea ice variables are bounded and because attainable sea ice states are limited to those in which stresses are (sub)critical. Studies have tried to overcome these limitations by carrying out the DA in the latent space of a variational autoencoder (VAE) in which the prior distribution is, by construction, approximately Gaussian. Problem with these approaches is that they either do not account for the time-variability of the prior distribution or require a recurrent network to do so. In addition to this, the observation operator and observational error covariances required by the DA are not available in the latent space. In this work we will show that these problems can be overcome by using two, online-trained VAEs: one for the model states and one for the differences between observed and modelled values (innovations). Once these are available a modified version of the ensemble transform Kalman filter (ETKF) can be used to calculate the optimal corrections in the two latent spaces without the need for observational error covariance. After this the corrected ensemble can be resampled using the VAE and moved forward in time using a dynamic model.

ABSTRACT. The ever-growing sector of wind energy underscores the importance of optimizing turbine operations and ensuring their maintenance with early fault detection mechanisms. Existing empirical and physics-based models provide approximate predictions of the generated power as a function of the wind speed, but face limitations in capturing the non-linear and complex relationships between input variables and output power. Data-driven methods present new avenues for enhancing wind turbine modeling using large datasets, thereby improving accuracy and efficiency. In this study, we use a hybrid semi-parametric model to leverage the strengths of two distinct approaches in a dataset with four turbines of a wind farm. Our model comprises a physics-inspired submodel, which offers a reliable approximation of the power, combined with a non-parametric submodel to predict the residual component. This non-parametric submodel is fed with a broader set of variables, aiming to capture phenomena not addressed by the physics-based part. For explainability purposes, the influence of input features on the output of the residual submodel is analyzed using SHAP values. The proposed hybrid model finally yields a 35-40\,\% accuracy improvement in the prediction of power generation. At the same time, the explainability analysis, along with the physics grounding from the parametric submodel, ensure deep understanding of the analyzed problem. In the end, this investigation paves the way for assessing the impact, and thus the potential optimization, of several unmodeled independent variables on the power generated by wind turbines.

ABSTRACT. In this paper we present a novel scientific machine learning reinterpretation of the well-known Kalman Filter, we explain its flexibil- ity in dealing with partially-unknown models and show its effectiveness in a couple of situations where the classic Kalman Filter is problematic.

ABSTRACT. Clustering climate models is crucial for identifying models satisfying desirable properties, such as the ability to ensure consistency with empirical observations and meaningful long-term climate dynamics [1,2,3]. The literature has largely addressed the issue of models that are too hot [3,4], in the sense that they project anomalously high temperatures ("hot model bias"). Models that are "too cold" have instead received considerably less attention. Reliable estimates of future temperatures, however, require both types of deviations to be properly addressed, as model averaging and similar approaches can be affected by upward or downward tilting of long-term temperature projections. Conventional clustering methods rely on spatially averaged models, thereby reducing the data to time series used for distance-based clustering [3]. In this paper, we propose an alternative approach to circumvent the information loss induced by spatial aggregation. We use a tensor representation of the climate models' dynamical systems and apply a higher-order clustering method [5] for unsupervised learning classification. The methodology demonstrates the ability to effectively distinguish between cold, warm, and hot models, thereby ensuring a closer alignment with commonly used climate sensitivity metrics without requiring ad-hoc adjustments [6]. We also demonstrate the economic value of the climate model selection approach by applying the unsupervised learning methodology to the quantification of the economic losses caused by climate change [7,8,9].

References: [1] Scafetta, N. (2022). Advanced testing of low, medium, and high ecs cmip6 gcm simulations versus era5-t2m. Geophysical Research Letters, 49(6):e2022GL097716. [2] Biffis, E. and Wang, S (2022). Downscaling of physical risks for climate scenario design. Singapore Green Finance Centre. [3] Biffis, E., Brandi, G. and Wang, S (2023). Unsupervised climate model selection: Assessing the hot model problem. Working paper, Imperial College Business School. [4] Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W., and Zelinka, M. (2022). Climate simulations: Recognize the ‘hot model ’ problem. [5] Brandi, G. and Di Matteo, T. (2021). Higher-Order Hierarchical Spectral Clustering for Multidimensional Data. In International Conference on Computational Science (pp. 387-400). Cham: Springer International Publishing. [6] IPCC (2023). The Earth’s Energy Budget, Climate Feedbacks and Climate Sensitivity, page 923–1054. Cambridge University Press. [7] Nordhaus, W. (2013). The climate casino: Risk, uncertainty, and economics for a warming world. Yale University Press. [8] Weitzman, M. L. (2012). Ghg targets as insurance against catastrophic climate damages. Journal of Public Economic Theory, 14(2):221–244. [9] Dietz, S. and Stern, N. (2015). Endogenous growth, convexity of damage and climate risk: how Nordhaus’ framework supports deep cuts in carbon emissions. The Economic Journal, 125(583):574–620.