Cluster Prototypes Research Articles

VLA Resolves Unexpected Radio Structures in the Perseus Cluster of Galaxies

Abstract We present new deep, high-resolution, 1.5 GHz observations of the prototypical nearby Perseus galaxy cluster from the Karl G. Jansky Very Large Array. We isolate for the first time the complete tail of radio emission of the bent-jet radio galaxy NGC 1272, which had been previously mistaken to be part of the radio mini-halo. The possibility that diffuse radio galaxy emission contributes to mini-halo emission may be a general phenomenon in relaxed cool-core clusters, and should be explored. The collimated jets of NGC 1272 initially bend to the west, and then transition eastward into faint, 60 kpc long extensions with eddy-like structures and filaments. We suggest interpretations for these structures that involve bulk motions of intracluster gas, the galaxy’s orbit in the cluster including projection effects, and the passage of the galaxy through a sloshing cold front. Instabilities and turbulence created at the surface of this cold front and in the turbulent wake of the infalling host galaxy most likely play a role in the formation of the observed structures. We also discover a series of faint rings, southeast of NGC 1272, which are a type of structure that has never been seen before in galaxy clusters.

Multi-task continuous learning model

In continual learning, previous learned knowledge tends to be overlapped by the subsequent training tasks. This bottleneck, known as catastrophic forgetting (CF), has recently been relieved between vision tasks involving simple tasks. Nevertheless, the challenge lies in the continuous classification of the sequential sets discriminated by global transformations, such as excessively spatial rotations. Aiming at this, a novel strategy of dynamic memory routing is proposed to dominate the forward paths of capsule network (CapsNet) according to the current input sets. To recall previous knowledge, a binary routing table is maintained among these sequential tasks. Then, an increment procedure of competitive prototype clustering is integrated to update the routing of current task. Moreover, a sparsity measurement is employed to decouple the salient routing among the different learned tasks. The experimental results demonstrate the superiority of the proposed memory network over the state-of-art approaches by the recalling evaluations on extended sets of SVHN, Cifar-100 and CelebA etc.

Multiple Stellar Populations of Globular Clusters from Homogeneous Ca–CN–CH–NH Photometry. VI. M3 (NGC 5272) Is Not a Prototypical Normal Globular Cluster* †

Abstract We present Ca–CN–CH–NH photometry for the well-known globular cluster (GC) M3 (NGC 5272). We show new evidence for two M3 populations with distinctly different carbon and nitrogen abundances, seen in a sharp division between CN-weak and CN-strong red-giant branches (RGBs) in M3. The CN-strong population shows a C–N anticorrelation that is a natural consequence of the CN cycle, while the CN-weak population shows at most a very weak C–N anticorrelation. Additionally, the CN-weak population exhibits an elongated spatial distribution that is likely linked to its fast rotation. Our derived metallicities reveal bimodal distributions in both populations, with 〈[Fe/H]〉 ≈ −1.60 and −1.45, which appear to be responsible for the discrete double RGB bumps in the CN-weak population and the large range. From this discovery, we propose that M3 consists of two GCs, namely C1 (23%, 〈[Fe/H]〉 ≈ −1.60) and C2 (77%, 〈[Fe/H]〉 ≈ −1.45), each of which has its own C–N anticorrelation and structural and kinematical properties, which are strong indications of independent systems in M3. The fractions of the CN-weak population for both C1 and C2 are high compared to Galactic GCs but they are in good agreement with GCs in the Magellanic Clouds. We suggest that M3 is a merger remnant of two GCs, most likely in a dwarf galaxy environment, and accreted to our Galaxy later in time. This is consistent with recent proposals of an ex situ origin for M3.

Distributive/integrative negotiation strategies in cross-cultural contexts: a comparative study of the USA and Italy

AbstractIntegrative and distributive negotiation strategies are a key paradigm of practice, teaching, and research. Are these US-formulated negotiation prototypes valid in the rest of the world? Adopting a cross-cultural view, we analyze a sample of 214 foreigners who detailed the negotiation behavior they faced in Italy (134) and in the United States (80). Implementing latent class analysis, we identify three clusters of negotiation prototypes. Our findings show how the Country is a predictor for cluster membership, and peculiar cultural traits of the two groups contribute to explain the differences in negotiation strategies. Three prototypes emerged: a typically distributive, an emotional integrative (mostly Italian), and an impersonal integrative (mostly American). Results show how the handling of emotions is a crucial part of the interaction for Italian negotiators, regardless of their orientation toward negotiation strategies, implying a cultural influence toward handling emotions in negotiations.

Fuzzy Centroid and Genetic Algorithms: Solutions for Numeric and Categorical Mixed Data Clustering

Abstract Statistical data analysis in machine learning and data mining usually uses the clustering technique. However, data with both attributes or mixed data exists universally in real life. K-prototype is a well-known algorithm for clustering mixed data because of its effectiveness in handling large data. However, practically, k-prototype has two main weaknesses, the use of mode as a cluster center for categorical attributes cannot accurately represent the objects, and the algorithm may stop at the local optimum solution because affected by random initial cluster prototypes. To overcome the first weakness, we can use fuzzy centroid, and for second weakness is to implement the genetic algorithm to search the global optimum solution. Our research combines the genetic algorithm and Fuzzy K-Prototype to accommodate these two weaknesses. We set up two multivariate data with high correlation and low correlation to see the robustness of the proposed algorithm. According to four value indexes of clustering result evaluation, Coefficient Varians Index, Partition Coefficient, Partition Entropy, and Purity, show that our proposed algorithm has a better result than K prototype. Based on the evaluation result, we conclude that our proposed algorithm can solve two weaknesses of the k-prototype algorithm.

Cognitive Matching of Design Subjects in Product Form Evolutionary Design.

In actual product development, the cognitive differences between users and designers make it difficult for the designed products to be recognized by users. To reduce the cognitive differences between these two design subjects, this paper proposes a method of cognitive matching of the design subjects. First, we use the relevant methods of Kansei engineering to quantify the Kansei image cognition of the two design subjects and construct a cognitive matching model of the design subjects with information entropy and the technique for order preference by similarity to ideal solution (TOPSIS). Second, according to the Kansei image, the Kansei image prototype cluster is constructed, and the representative Kansei image prototype is obtained. Then, we combine an artificial neural network (ANN) with a cognitive matching model of the design subjects to construct a product Kansei image evaluation system; this is used to evaluate the evolved forms. Finally, a product Kansei image form evolution system is constructed based on the genetic algorithm (GA). To some extent, the system simulates the cognitive matching process between designers and users in product design, helps designers to more accurately understand the cognitive trends of the two design subjects, and provides a theoretical basis for the intelligent design of product forms through the cognitive balance of multiple design subjects. This paper takes a beverage bottle as an example to verify the feasibility of the model through a comparative study.

Application of FCMAlgorithm Combined with Artcial Neural Network in TBM Operation Data

Fuzzy clustering theory is widely used in data mining of full-face tunnel boring machine. However, the traditional fuzzy clustering algorithm based on objective function is difficult to effectively cluster functional data. We propose a new Fuzzy clustering algorithm, namely FCM–ANN algorithm. The algorithm replaces the clustering prototype of the FCM algorithm with the predicted value of the articial neural network. This makes the algorithm not only satisfy the clustering based on the traditional similarity criterion, but also can effectively cluster the functional data. In this paper, we rst use the t-test as an evaluation index and apply the FCM–ANN algorithm to the synthetic datasets for validity testing. Then the algorithm is applied to TBM operation data and combined with the crossvalidation method to predict the tunneling speed. The predicted results are evaluated by RMSE and R2 . According to the experimental results on the synthetic datasets, we obtain the relationship among the membership threshold, the number of samples, the number of attributes and the noise. Accordingly, the datasets can be effectively adjusted. Applying the FCM–ANN algorithm to the TBM operation data can accurately predict the tunneling speed. The FCM–ANN algorithm has improved the traditional fuzzy clustering algorithm, which can be used not only for the prediction of tunneling speed of TBM but also for clustering or prediction of other functional data.

Convex clustering method for compositional data via sparse group lasso

High-dimensional sparse clustering with compositional data is of great practical importance, as exemplified by applications in high-throughput gene expression profiles analysis. In this paper, we develop a compositional clustering framework based on convex clustering, which is a convex relaxation of hierarchical clustering that incorporates a fused penalty term on the cluster prototypes. To explicitly deal with the issue of high dimensionality and sparsity, we propose the Compositional Convex Clustering with Sparse Group Lasso (CCC-SGL). The isometric logratio (ilr) transformation is first applied to transform the composition in the simplex space to the standard Euclidean geometry. Then, a group lasso penalty and a lasso penalty are imposed on the cluster centers, which effectively selects informative features and promotes within-feature sparsity. The proposed convex clustering formulation is numerically and efficiently solved with the proximal gradient descent algorithm within the Alternating Direction Method of Multipliers (ADMM) framework. Simulation studies are carried out to evaluate the performance of the proposed methodology and also a real data set in microbiome sequencing is analyzed.

Multiscale dynamic fusion prototypical cluster network for fault diagnosis of planetary gearbox under few labeled samples

Deep learning has been widely applied for fault diagnosis of planetary gearbox because of its powerful feature extraction and nonlinear capabilities. However, due to the limited labeled samples of planetary gearbox in some industrial application scenarios, traditional deep learning methods based on sufficient labeled data cannot achieve satisfactory results. In this work, a novel deep learning method named multiscale dynamic fusion prototypical cluster network (MFPCN) is proposed to address this issue. Firstly, multiscale dynamic fusion module is designed to dynamically modulate multiscale features through a gate mechanism for obtaining more discriminative features under few labeled samples. After that, prototype fuzzy c-means cluster algorithm is proposed to refine the position of class prototypes in high-dimensional space by merging unlabeled samples information, making it more representative of the same class features, which provides a more accurate distance metric benchmark for prototypical nearest neighbor classifier. Finally, joint learning scheme is constructed by combining prototypical nearest neighbor classifier and global classifier to jointly train the entire network. Extensive experiments and the engineering application show that the proposed method is more effective in the case of few labeled samples than other methods.

Data clustering via cooperative games: A novel approach and comparative study

Arguably, the main purpose of cluster analysis is to develop algorithms to reveal natural groupings (clusterings) over a set of data points based on their similarity. On the other hand, the focus of cooperative game theory (CGT) is to study the formation of groups (coalitions) of decision makers (players) and ways to split the resulting income among them. Due to the conceptual similitude between these fields, algorithms rooted in CGT have recently emerged for tackling the data clustering problem. In this work, we revisit two such algorithms, one based on cluster prototypes (Biobjective Game Clustering – BiGC) and the other based on dense regions of data points (Density-Restricted Agglomerative Clustering – DRAC). We also present a novel partitional clustering algorithm, referred to as HGC (after Hedonic Game based Clustering), which is grounded on theoretical results stemming from the subclass of hedonic games. Two HGC versions are investigated, which differ in the order of the players in the game, and a detailed factorial simulation study is reported to analyze how sensitive they are to three relevant factors, namely number of clusters, number of features, and noise level. Besides, a heuristic to calibrate the value of HGC’s single control parameter (viz., the number of nearest neighbors of each point) is provided, so as to yield high-quality partitions. To compare the performance of the CGT algorithms, a series of experiments were conducted on UCI and gene-expression data sets, the majority of which being high dimensional. Overall, the results measured by 10 external validation indices evidence that HGC is usually more stable and effective than DRAC and BiGC. They also show that HGC is very competitive (sometimes, considerably better) to well-known clustering algorithms/variants (specifically, k-means, k-means++, affinity propagation, two variants of hierarchical clustering, and the density peak clustering algorithm). Remarkably, HGC could fully recover the true clustering structures for two gene-expression data sets.

Identification of Fuzzy Rule-Based Models With Output Space Knowledge Guidance

In this article, we advocate that a knowledge tidbit residing in the output space could be helpful in improving the performance (accuracy) of the fuzzy rule-based model. It states that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">if two outputs are far apart from each other</i> , <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">it is advisable to place their corresponding inputs in different clusters when forming subspaces of the input space</i> . Considering this knowledge guidance mechanism, we propose two different methods to partition the input space. In the first method, input data are first partitioned with the use of the standard clustering algorithm, say fuzzy C-means; here, a constructed partition matrix is reflective of the structure present in the input space. Then, the knowledge tidbit is used to adjust the entries of the original partition matrix in such a way that those input data whose corresponding output data are far apart from each other are assigned with low values of proximity. In the second method, we propose two strategies to modify the distance between input data and a prototype (cluster center) identified in the input space. The crux of this method is that if there are many input data (which, in virtue of the knowledge tidbit, are regarded as being far-apart from the input data of interest) around a certain prototype, the distance between the input data of interest and this prototype should be penalized. Thus, the membership of these input data to the prototype is reduced. The comprehensive experimental studies carried out on both synthetic and publicly available data are used to examine the usefulness of the proposed methods.

Artificial immune network clustering based on a cultural algorithm

Data mining technology has been applied in many fields. Prototype-based cluster analysis is an important data mining method, but its ability to discover knowledge is limited because of the need to know the number of target data categories and cluster prototypes in advance. Artificial immune evolutionary network clustering is a clustering method based on network structure. Compared with prototype-based cluster analysis, it has the advantage of realizing unsupervised learning and clustering without any prior knowledge of data. However, artificial immune evolutionary network clustering also has problems such as a lack of guidance in the clustering process, fuzzy boundary sensitivity, and difficulty in determining parameters. To solve these problems, an artificial immune network clustering algorithm based on a cultural algorithm is proposed. First, three kinds of knowledge are constructed: normative knowledge is used to regulate the spatial range of population initialization to avoid blindness; state knowledge is used to distinguish the type of antigen, and immune defense measures are taken to prevent the network structure caused by noise and boundaries from being unclear; and topology knowledge is used to guide the antigen for optimal antibody search. Second, topology knowledge in the cultural algorithm is used to characterize the distribution of antigens and antibodies in space, and elite learning is used to improve the traditional clone mutation operator. Based on the shadow set theory, a method for adaptively determining the compression threshold is proposed. Finally, the results of simulation experiments show that the proposed algorithm can effectively overcome the above problems, and the clustering performances on a synthetic dataset and an actual dataset are satisfactory.

Multi-epoch X-ray imaging of globular cluster M62 with Chandra

ABSTRACT Using archival spectral-imaging data with a total exposure of ∼144 ks obtained by Chandra, 43 X-ray sources are detected within the half-light radius of globular cluster M62 (NGC 6266). Based on the X-ray colour–luminosity diagram or the positional coincidences with known sources, we have classified these sources into different groups of compact binaries including cataclysmic variable (CV), quiescent low-mass X-ray binary (qLMXB), millisecond pulsar, and black hole (BH). Candidates of the X-ray counterparts of 12 CVs, 4 qLMXBs, 2 MSPs, and 1 BH are identified in our analysis. The data used in our analysis consist of two frames separated by 12 yr, which enable us to search for the long-term variability as well as the short-term X-ray flux variability within each observation window. Evidence for the short-term variability and long-term variability have been found in 7 and 12 sources, respectively. For a number of bright sources with X-ray luminosities Lx ≳ 1032 erg s−1, we have characterized their spectral properties in further details. By comparing the X-ray population in M62 with those in several other prototypical globular clusters, we found the proportion of bright sources is larger in M62 that can possibly be a result of their active dynamical formation processes.

Formulation of a Dressed Coupled-Cluster Method with Implicit Triple Excitations and Benchmark Application to Hydrogen-Bonded Systems.

In this paper, we present a coupled-cluster theory based on a double-exponential wave operator ansatz, which is capable of mimicking the effects of connected triple excitations in an iterative manner. The triply excited manifold is spanned via the action of a set of scattering operators on doubly excited determinants, whereas their action annihilates the Hartree-Fock reference determinant. The effect of triple excitations is included at a computational scaling slightly higher than that of conventional coupled-cluster singles and doubles. Furthermore, we demonstrate two approximate schemes, which arise naturally, and argue that both these schemes come equipped with certain renormalization terms capable of handling nonbonding interactions due to robust inclusion of the screened Coulomb interaction. We justify our claims from both a theoretical perspective and a number of numerical applications to prototypical water clusters, in a number of basis functions. Our methods show overall comparable performance to the canonical coupled-cluster theory with singles, doubles, and perturbative triples (CCSD(T)) and allied methods, however, at a lower computational scaling.

Lumbopelvic Rhythm Prediction Using Machine Learning And Its Use As A Biomarker For Low Back Pain Identification

Lumbopelvic (LUM-PEL) rhythm has the potential be used as biomarker for diagnosis and rehabilitation of athletes predisposed to mechanical low back pain (LBP). Studies till date have mostly focused on discrete variables from the time series to explain movement patterns. Machine learning algorithms provide opportunity to analyze continuous time series data for predictive classification of movement patterns into pathological and non-pathological, adding value to early diagnosis and clinical decision making for conditions such as LBP. PURPOSE: Use of machine learning to categorize healthy LUM-PEL rhythm. METHODS: 79 participants with no LBP (Young: n=42; 18-40yr; 27.6±6.5yr; Older: n=37; 41-65yr; 51.7±7.3yr). 3D segmental kinematics of lumbar (LUM: L1-L5) and pelvis [PEL] were calculated for maximum trunk flexion-extension. Coordination patterns were divided into in-phase, anti-phase, superior and inferior-only based on the coupling angles of LUM and PEL. K-means clustering, an unsupervised machine learning algorithm, was employed to create clusters of movement patterns of the coupling and segmental angles based on dynamic time warping similarity. Sample distribution within each cluster was compared for different age groups. RESULTS: LUM-PEL rhythm fell under k=3 major movement pattern clusters (Fig. 1). No difference between age groups was observed. Non-pathological LUM-PEL rhythm clusters suggest flexion movement initiation and return from hyperextension typically have segments in anti-phase (LUM leading: 40.4%), PEL/ LUM only (35.3%) and in-phase (LUM leading: 24.3%). The 2 segments predominantly move in-phase except at start and end of movement. Patterns were not apparent when using segment angles or through discrete variable of mean coupling angles.Figure 1:: Prototypical Movement Pattern Clusters of Lumbopelvic Coupling Angles During Flexion/ExtensionCONCLUSIONS: Using the discovered movement pattern clusters, individuals with LBP could be identified and training prescriptions can be based on healthy segmental coordination.

Outlier Detection in Growth Data: Beyond Biologically Implausible Values

Abstract Objectives Eliminating anthropometry measurement error and employing outlier and biological implausible values (BIV) detection methods adapted to longitudinal measurements is important for the study of growth. This work aimed to review and assess the accuracy of the available BIV and outlier detection methods and propose a growth trajectory outlier detection method. Methods We included 2354 infants from the Applied Research Group for Kids (TARGet Kids! ) cohort-based in Toronto (ON, Canada) that recruits healthy children from birth to 5 years of age. We considered infants with at least 8 length and weight measurements available between the 1st and the 24th month of age. Weight-for-length z-scores (wflz) were calculated using the WHO growth standards. Outlier measurements were randomly introduced in 5% of the wflz measurements using a normal distribution (μ = 0, σ = 1). We employed 4 outlier detection methods; an empirical detection method for BIV using the cut-offs derived from the WHO Child Growth Standards, a clustering method, a method based on cluster prototypes for individual outlier measurements and a method based on cluster prototypes for entire growth trajectories. Each method was applied individually and evaluated using the sensitivity and specificity indexes based on the manually introduced outliers. We also calculated the Kappa statistic to evaluate the agreement of each method against the manual outliers. Results After excluding premature (&amp;lt;37 weeks), low birth weight (&amp;lt;1500 g) neonates and children with missing length and weight measurements, we analyzed 393 children with a total of 3144 measurements. Sensitivity and specificity for the four methods ranged between 4.4%–55.0% and 83.7% −99.7%, respectively, with kappa being non-significant (P &amp;gt; 0.05) only for the empirical. The clustering detection method reported a higher finding rate, while the empirical method found most of the BIV, but few of the rest of the outliers. Conclusions BIV account for a small portion of the possible outliers in growth datasets. We show that additional statistical or model-based methods are required for a more comprehensive outlier detection process, which has implications for growth analysis and nutritional assessment. Funding Sources Joannah and Brian Lawson Center for Child Nutrition, Connaught Fund, Onassis Foundation.

Semi-Supervised Multi-view clustering based on orthonormality-constrained nonnegative matrix factorization

Multi-view clustering aims at integrating the complementary information between different views so as to obtain an accurate clustering result. In addition, the traditional clustering is a kind of unsupervised learning method, which does not take the label information into learning. In this paper, we propose a novel model, called semi-supervised multi-view clustering based on orthonormality-constrained nonnegative matrix factorization (MVOCNMF), to cluster the multi-view data into a number of categories. In the proposed model, based on the label information, we first learn the low-dimensional representations of data by the constrained NMF technique, and simultaneously cluster the samples with the same label into the clustering prototypes for each view. After that, we put forward a novel orthonormality constraint term to obtain the desirable representations for each view, and use the co-regularization to integrate the complementary information from different views. We further develop an alternating minimization algorithm to solve the proposed model, and present the convergence analysis and computational complexity of the proposed method. Extensive experimental results on several multi-view datasets have shown that the proposed MVOCNMF method outperforms the existing multi-view clustering methods.

Higher- and lower-order personality traits and cluster subtypes in social anxiety disorder.

Social anxiety disorder (SAD) can come in different forms, presenting problems for diagnostic classification. Here, we examined personality traits in a large sample of patients (N = 265) diagnosed with SAD in comparison to healthy controls (N = 164) by use of the Revised NEO Personality Inventory (NEO-PI-R) and Karolinska Scales of Personality (KSP). In addition, we identified subtypes of SAD based on cluster analysis of the NEO-PI-R Big Five personality dimensions. Significant group differences in personality traits between patients and controls were noted on all Big Five dimensions except agreeableness. Group differences were further noted on most lower-order facets of NEO-PI-R, and nearly all KSP variables. A logistic regression analysis showed, however, that only neuroticism and extraversion remained significant independent predictors of patient/control group when controlling for the effects of the other Big Five dimensions. Also, only neuroticism and extraversion yielded large effect sizes when SAD patients were compared to Swedish normative data for the NEO-PI-R. A two-step cluster analysis resulted in three separate clusters labelled Prototypical (33%), Introvert-Conscientious (29%), and Instable-Open (38%) SAD. Individuals in the Prototypical cluster deviated most on the Big Five dimensions and they were at the most severe end in profile analyses of social anxiety, self-rated fear during public speaking, trait anxiety, and anxiety-related KSP variables. While additional studies are needed to determine if personality subtypes in SAD differ in etiological and treatment-related factors, the present results demonstrate considerable personality heterogeneity in socially anxious individuals, further underscoring that SAD is a multidimensional disorder.

A novel fuzzy clustering algorithm by minimizing global and spatially constrained likelihood-based local entropies for noisy 3D brain MR image segmentation

In this paper, we propose a novel fuzzy clustering algorithm by minimizing global and spatially constrained likelihood-based local entropies (FCMGsLE) for segmenting noisy 3D brain magnetic resonance (MR) image volumes. For each voxel, in order to measure uncertainties that arise while identifying its class, two different entropies are defined. In particular, they measure the amount of uncertainties in terms of global entropy using fuzzifier weighted global membership function and spatially constrained likelihood-based local entropy using fuzzifier weighted local membership function. To mitigate the effect of noise and intensity inhomogeneity (IIH) or radio frequency (RF) inhomogeneity, the local membership function is induced by spatially constrained likelihood measure. These entropies are minimized through a fuzzy objective function to obtain the cluster prototypes and membership functions. The final membership function is obtained by integrating these global and local membership functions using weighted parameters. The algorithm is assessed both qualitatively and quantitatively on ten 3D volumes of simulated and clinical brain MR image data having high levels of noise and intensity inhomogeneity and a synthetic 3D image volume with Rician noise. The simulation results reveal that the proposed algorithm outperforms several state-of-the-art algorithms devised in recent past when evaluated in terms of segmentation accuracy, Dice similarity coefficient, partition coefficient, and partition entropy

Managing the CERN Batch System with Kubernetes

The CERN Batch Service faces many challenges in order to get ready for the computing demands of future LHC runs. These challenges require that we look at all potential resources, assessing how efficiently we use them and that we explore different alternatives to exploit opportunistic resources in our infrastructure as well as outside of the CERN computing centre. Several projects, like BEER, Helix Nebula Science Cloud and the new OCRE project, have proven our ability to run batch workloads on a wide range of non-traditional resources. However, the challenge is not only to obtain the raw compute resources needed but how to define an operational model that is cost and time efficient, scalable and flexible enough to adapt to a heterogeneous infrastructure. In order to tackle both the provisioning and operational challenges it was decided to use Kubernetes. By using Kubernetes we benefit from a de-facto standard in containerised environments, available in nearly all cloud providers and surrounded by a vibrant ecosystem of open-source projects. Leveraging Kubernetes’ built-in functionality, and other open-source tools such as Helm, Terraform and GitLab CI, we have deployed a first cluster prototype which we discuss in detail. The effort has simplified many of the existing operational procedures we currently have, but has also made us rethink established procedures and assumptions that were only valid in a VM-based cloud environment. This contribution presents how we have adopted Kubernetes into the CERN Batch Service, the impact its adoption has in daily operations, a comparison on resource usage efficiency and the experience so far evolving our infrastructure towards this model.