Computing Support Research Articles

Quantum support vector machines: theory and applications

Abstract. Quantum Support Vector Machines (QSVMs) combine the fundamental principles of quantum computing and classical Support Vector Machines (SVMs) to improve machine learning performance. In this paper, the author will further explore QSVM. Firstly, introduce the basics of classical SVM, including hyperplane, margin, support vector, and kernel methods. Then, introduce the basic theories of quantum computing, including quantum bits, entanglement, quantum states, superposition, and some related quantum algorithms. Focuses on the concept of QSVM, quantum kernel methods, and how SVM runs on a quantum computer. Key topics include quantum state preparation, measurement, and output interpretation. Theoretical advantages of QSVMs, such as faster computation speed, stronger performance to process high-dimensional data, and kernal computation. In addition, the author discussed the implementation of QSVM, quantum algorithms, quantum gradient descent, and optimization techniques for SVM training. The article also discusses practical issues such as error mitigation and quantum hardware requirements. The purpose of this paper is to show the advantages of QSVM by comparing SVM and QSVM.

FLARE: field line analysis and reconstruction for 3D boundary plasma modeling

Abstract The FLARE code is a magnetic mesh generator that is integrated within a suite of tools for the analysis of the magnetic geometry in toroidal fusion devices. A magnetic mesh is constructed from field line segments and permits fast reconstruction of field lines in 3D boundary plasma codes such as EMC3-EIRENE. Both intrinsically non-axisymmetric configurations (stellarators) and those with symmetry breaking perturbations of an axisymmetric equilibrium (tokamaks) are supported. The code itself is written in Modern Fortran with MPI support for parallel computing, and it incorporates object-oriented programming for the definition of the magnetic field and the material surface geometry. Extended derived types for a number of different magnetohydrodynamic equilibrium and plasma response models are implemented. The core element of FLARE is a field line tracer with adaptive step-size control, and this is integrated into tools for the construction of Poincaré maps and invariant manifolds of X-points. A collection of high-level procedures that generate output files for visualization is build on top of that. The analysis modules are build with Python frontends that facilitate customization of tasks and/or scripting of parameter scans.

PyTWMR: transcriptome-wide Mendelian randomization in python.

Mendelian randomization (MR) is a widely used approach to estimate causal effect of variation in gene expression on complex traits. Among several MR-based algorithms, transcriptome-wide summary statistics-based Mendelian Randomization approach (TWMR) enables the uses of multiple SNPs as instruments and multiple gene expression traits as exposures to facilitate causal inference in observational studies. Here we present a Python-based implementation of TWMR and revTWMR. Our implementation offers GPU computational support for faster computations and robust computation mode resilient to highly correlated gene expressions and genetic variants. pyTWMR is available at github.com/soreshkov/pyTWMR.

Development of a multi-layer canopy model for E3SM Land Model with support for heterogeneous computing

The vertical structure of vegetation canopies creates micro-climates. However, the land components of most Earth System Models, including the Energy Exascale Earth System Model (E3SM), typically neglect vertical canopy structure by using a single layer big-leaf representation to simulate water, CO2, and energy exchanges between the land and the atmosphere. In this study, we developed a Multi-Layer Canopy Model for the E3SM Land Model to resolve the micro-climate created by vegetation canopies. The model developed in this study re-implements the CLM-ml_v1 to support heterogeneous computing architectures consisting of CPUs and GPUs and includes three additional optimization-based stomatal conductance models. The use of Portable, Extensible Toolkit for Scientific Computation provides a speedup of 25–50 times on a GPU relative to a CPU. The numerical implementation of the model was verified against CLM-ml_v1 for a month-long simulation using data from the Ameriflux US-University of Michigan Biological Station site. Model structural uncertainty was explored by performing control simulations for five stomatal conductance models that exclude and include the control of plant hydrodynamics (PHD) on photosynthesis. The bias in simulated sensible and latent heat fluxes was lower when PHD was accounted for in the model. Additionally, six idealized simulations were performed to study the impact of three environmental variables (i.e. air temperature, atmospheric CO2, and soil moisture) on canopy processes (i.e. net CO2 assimilation, leaf temperature, and leaf water potential). Increasing air temperature reduced net CO2 assimilation and increased air temperature. Net CO2 assimilation increased at higher atmospheric CO2, while decreasing soil moisture resulted in lower leaf water potential.

Testing the Resilience of MEC-Based IoT Applications Against Resource Exhaustion Attacks

Multi-access Edge Computing (MEC) is an emerging computing model that provides the necessary on-demand resources and services to the edge of the network, ensuring powerful computing, storage capacity, mobility, location, and context awareness support to emerging Internet of Things (IoT) applications. Nonetheless, its complex hierarchical model introduces new architectural interdependencies, which can influence the resilience of IoT applications against cyber attacks. Although application resilience has been investigated in the context of cloud computing, existing studies are not directly applicable to such an extended edge-cloud paradigm. The use of different enabling technologies at the edge of the network, such as various wireless access technologies and virtualization, implies several threats and challenges that make the analysis and deployment of resilience mechanisms a technically challenging problem. In this paper, we first present an overview of the threat model, describing the threats for the different layers of this paradigm. We then study the impact of resource-exhausting attacks – a particularly relevant class for this paradigm - on three different IoT applications exploiting the services offered by the MEC-based architecture. We adopt a testing-based methodology conceived to characterize the resilience of such applications under attack. A set of most important resilience-related indicators are also identified. The characterization's results are useful to support the analyst in planning proper protection means at individual architectural layers.

An Investigation on Utilizing Large Language Model for Industrial Computer-Aided Design Automation

Text-to-image or 3D content generation has attracted attention in academia and industry because of its impressive performance in high-fidelity content generation with natural language commands. However, its widespread application in industrial computation-aided design (CAD) is not yet found. The main issue lies in the end-to-end generation through implicit probabilistic learning from datasets. This presents difficulties in precise, controllable, and transparent creations. Also, it is time-consuming and labor-intensive for designers to verify and customize the generated result to fit their demands. Furthermore, popular generative models offer limited support for early-stage engineering computations, leaving tedious and massive calculations to human engineers. To overcome these challenges, this paper proposes a method to apply large language model (LLM) to assist standard CAD design workflow on both engineering computation and precise 3D geometry generation. During the designing process, LLM can support the computation process with human guidance and generate accurate 3D shapes with the capability of sequential code representation and transformation. To verify the feasibility of the proposed method, a use case for a one-stage reduction gear system consisting of two gears will be given as a proof-of-concept demonstration. The result shows that the LLM can carry out the computation and get the result as expected with only a few lines of human guidance and feedback. Then, it can create CAD models with accurate parameters aligned to the computational results. It can cope with various requirements given by users and allow users to further modify the generation using the same CAD environment without much effort. Authors believe that this piece of work can shed light on utilizing the capability of LLMs for a certain part of task automation based on human beings’ intentions. With the rise of LLMs and relevant technologies, conventional CAD may shift to a more user-friendly AI-aided design (AIAD) diagram.

Design of Multi-Competitors Winner-Take-All Neural Networks Based on DNA Strand Displacement for Molecular Pattern Recognition

DNA strand displacement technology (DSDT) provides flexible and powerful technical support for DNA molecular computing. DNA-based neural networks with Winner-Take-All (WTA) strategy has a great potential for nonlinear calculation. However, so far it has been limited to achieving the simultaneous competition of two competitors. Optimizing the calculation model and reducing system response time to recognize complex and functional molecular patterns remains a huge challenge. Here a novel neural network with WTA strategy based on DSDT was constructed, which allowed three competitors to participate in the competition at the same time. Firstly, the feasibility of the three-competitor WTA neural network was proved by 9-bit pattern recognition. Then the three-competitors WTA neural network was further extended to larger scale pattern recognition, which successfully recognized 64-bit letters A, B, and C and 100-bit handwritten digits 0, 2, and 4, respectively. Simulations showed that when recognizing the same target patterns with same number bits, compared with two-competitors WTA neural network, the three-competitors WTA network only used down to two-thirds DNA strands, and the system response time was reduced by more than ten times. This paper demonstrated the efficient recognition ability of the three-competitor WTA neural network, which is expected to be used to identify more complex information.

GPU-accelerated non-dominated sorting genetic algorithm III for maximizing protein production

<abstract><p>Maximizing protein expression levels poses a major challenge in bioengineering. To increase protein expression levels, numerous factors, including codon bias, codon context bias, hidden stop codons, homologous recombination, suitable guanine-cytosine ratio, and hairpin loop structure, are crucial and quantified by six objective functions: CAI, CPB, HSC, HD, GC3, and SL. Optimizing these six objectives simultaneously constitutes a multi-objective optimization problem, aiming to identify the favorable Pareto solutions rather than a singular optimal solution. However, achieving satisfactory solutions requires numerous cycles and solutions, thus leading to a large number of functional evaluations. While there are frameworks for multi-objective optimization problems, they often lack efficient support for objective function computation in protein encoding. In this paper, we proposed a method to design a set of coding sequences (CDSs) based on non-dominated sorting genetic algorithm III (NSGA-III), accelerated using NVIDIA graphical processing units (GPUs). Experimental results indicated that our method is 15,454 times faster than the Pymoo framework and is evaluated using 100 solutions and 100 cycles. Since our GPU implementation facilitated the use of larger solutions and more cycles, we were able to design a superior set of CDSs by increasing solutions to 400 and cycles to 12,800. In addition, our NSGA-III-based method consistently surpassed the NSGA-II approach when the number of cycles exceeded 3200 by utilizing 100 solutions. Finally, we observed that a gradual reduction of the mutation probability as the number of cycles increased yielded better quality results than maintaining a fixed mutation probability.</p></abstract>

DELOFF: Decentralized Learning-Based Task Offloading for Multi-UAVs in U2X-Assisted Heterogeneous Networks

With multi-sensors embedded, flexible unmanned aerial vehicles (UAVs) can collect sensory data and provide various services for all walks of life. However, limited computing capability and battery energy put a great burden on UAVs to handle emerging compute-intensive applications, necessitating them to resort to innovative computation offloading technique to guarantee quality of service. Existing research mainly focuses on solving the offloading problem under known global information, or applying centralized offloading frameworks when facing dynamic environments. Yet, the maneuverability of today’s UAVs, their large-scale clustering, and their increasing operation in the environment with unrevealed information pose huge challenges to previous work. In this paper, in order to enhance the long-term offloading performance and scalability for multi-UAVs, we develop a decentralized offloading scheme named DELOFF with the support of mobile edge computing (MEC). DELOFF considers the information uncertainty caused by the dynamic environment, uses UAV-to-everything (U2X)-assisted heterogeneous networks to extend network resources and offloading flexibility, and tackles the joint strategy making related to computation mode, network selection, and offloading allocation for multi-UAVs. Specifically, the optimization problem of multi-UAVs is addressed by the proposed offloading algorithm based on a multi-arm bandit learning model, where each UAV itself can adaptively assess the offloading link quality through the fuzzy logic-based pre-screening mechanism designed. The convergence and effectiveness of the DELOFF proposed are also demonstrated in simulations. And, the results confirm that DELOFF is superior to the four benchmarks in many respects, such as reduced consumed energy and delay in the task completion of UAVs.

Comparing performance of neural network frameworks for image recognition: A review

Spiking Neural Networks (SNNs) are increasingly recognized as a promising approach to simulating the biological behavior of neurons. This study conducted an in-depth performance comparison of different SNN models with supervised learning algorithms, particularly in the field of image recognition. Using datasets like MNIST, CIFAR-10, and ImageNet, the research analyzed the performance and accuracy of SNNs compared to traditional Artificial Neural Networks (ANNs). This research contributes to understanding SNNs' performance in supervised learning environments and offers insights into the optimization of SNN architectures, thus influencing future research and the development of next-generation neural network designs. Based on this comparative study, it was found that while SNNs perform admirably with smaller, less complex datasets such as MNIST and CIFAR, they encounter difficulties with larger, more complex datasets like ImageNet. Despite these challenges, advances in conversion techniques have led to models that more closely simulate the behavior of biological neurons. The study identified significant future work areas, including addressing issues of local learning methods, the limited scale of SNNs, hardware implementation difficulties, a lack of substantial benchmark datasets, and limited support for complex computations.

Mobile Edge Computing (MEC)-Enabled UAV Placement and Computation Efficiency Maximization in Disaster Scenario

Unmanned aerial vehicles (UAVs) are considered for disaster management to provide better coverage or help first responders provide advanced services. Furthermore, UAVs combined with mobile edge computing (MEC) capabilities, have recently emerged as a support for cloud computing by deploying UAV-assisted MEC at the network edge to reduce the latency and load of cloud computing data centers. We investigate the problem of maximizing the utility function by jointly optimizing the computing power, user associations, performance, duration, and location of user equipments (UEs) in UAV-assisted MEC networks in a disaster scenario. We considered several constraints including delay, quality of service, and coverage. The optimization problem is a mixed-integer non-linear programming problem. We propose a multi-stage offloading algorithm based on a learning algorithm and an interior-point method to obtain a viable solution. The simulation results obtained demonstrate the effectiveness of our proposed algorithm compared to existing schemes.

Redactable consortium blockchain based on verifiable distributed chameleon hash functions

With the evolving application demands, the inherent immutability of consortium blockchains hinders their widespread adoption. For example, expired data stored on the chain cannot be deleted, and erroneous data cannot be redacted, seriously limiting the flexibility of consortium blockchains. However, existing redactable blockchain solutions need to be improved in aspects of decentralization, efficiency, and fault tolerance. This paper develops a new verifiable distributed chameleon hash (VDCH) function to solve the above problems. With VDCH, nodes share chameleon keys with a secure multi-party computation protocol based on a verifiable key-sharing scheme, and the collision shares can be verified with a Schnorr non-interactive zero-knowledge proof protocol, which enhances the fault tolerance of the consortium chain while maintaining its decentralized nature. Then, this paper proposes a consensus protocol called CVTSS based on verifiable threshold signatures, which provides protocol support for collaborative hash collision computation by multiple nodes using VDCH, thus avoiding the dependence on Nakamoto consensus and improving the redaction efficiency. Meanwhile, CVTSS uses threshold signatures to prevent malicious nodes from tampering with data using one-time chameleon keys. Finally, this paper constructs an efficient, practical, and secure redactable consortium chain scheme based on VDCH and CVTSS. Theoretical analysis and experimental results show that the proposed scheme can operate safely in the presence of malicious nodes with an acceptable time cost.

Human-Computer Collaborative Visual Design Creation Assisted by Artificial Intelligence

With the support and promotion of big data and cloud computing, AI has penetrated into every field of people's lives more and more deeply, with its characteristics of sustainable work, extremely fast computing speed, and a certain intelligence. This is an effective way to solve the general lack of demand and productivity of visual design, and relieve the pressure off designers to deal with relatively low-quality and high-demand designs. Therefore, the combination of design and artificial intelligence technology is a necessity. Research on the application of artificial intelligence technology for visual design is also in full swing at home and abroad However, at present, teams at home and abroad are in the exploratory stage. This paper considers whether it is possible to build an intelligent visual design and creation system by using artificial intelligence technology to help graphic communication designers achieve high-quality, high-efficiency, and high-quantity design output. Additionally, this paperexplores how to combine artificial intelligence technology with designers' design workflow so as to form a complementary human-computer cooperation mode. We will explore how to integrate AI technology with designers' design workflow and then create a human-machine collaboration model with complementary advantages to achieve the high quality, high efficiency, and high quantity of design output required by the intelligent visual design creation system being built. Finally, a basic framework of a generative smart human-computer collaborative visual design creation system based on a subset of neural network expert systems in multiple domains and an aggregate of different modules supported by the system is formed, and the working principle and usage process of the system are further elaborated with the example of packaging design.

Overflow-free Compute Memories for Edge AI Acceleration

Compute memories are memory arrays augmented with dedicated logic to support arithmetic. They support the efficient execution of data-centric computing patterns, such as those characterizing Artificial Intelligence (AI) algorithms. These architectures can provide computing capabilities as part of the memory array structures (In-Memory Computing, IMC) or at their immediate periphery (Near-Memory Computing, NMC). By bringing the processing elements inside (or very close to) storage, compute memories minimize the cost of data access. Moreover, highly parallel (and, hence, high-performance) computations are enabled by exploiting the regular structure of memory arrays. However, the regular layout of memory elements also constrains the data range of inputs and outputs, since the bitwidths of operands and results stored at each address cannot be freely varied. Addressing this challenge, we herein propose a HW/SW co-design methodology combining careful per-layer quantization and inter-layer scaling with lightweight hardware support for overflow-free computation of dot-vector operations. We demonstrate their use to implement the convolutional and fully connected layers of AI models. We embody our strategy in two implementations, based on IMC and NMC, respectively. Experimental results highlight that an area overhead of only 10.5% (for IMC) and 12.9% (for NMC) is required when interfacing with a 2KB subarray. Furthermore, inferences on benchmark CNNs show negligible accuracy degradation due to quantization for equivalent floating-point implementations.

Network Resource Allocation Algorithm Using Reinforcement Learning Policy-Based Network in a Smart Grid Scenario

The exponential growth in user numbers has resulted in an overwhelming surge in data that the smart grid must process. To tackle this challenge, edge computing emerges as a vital solution. However, the current heuristic resource scheduling approaches often suffer from resource fragmentation and consequently get stuck in local optimum solutions. This paper introduces a novel network resource allocation method for multi-domain virtual networks with the support of edge computing. The approach entails modeling the edge network as a multi-domain virtual network model and formulating resource constraints specific to the edge computing network. Secondly, a policy network is constructed for reinforcement learning (RL) and an optimal resource allocation strategy is obtained under the premise of ensuring resource requirements. In the experimental section, our algorithm is compared with three other algorithms. The experimental results show that the algorithm has an average increase of 5.30%, 8.85%, 15.47% and 22.67% in long-term average revenue–cost ratio, virtual network request acceptance ratio, long-term average revenue and CPU resource utilization, respectively.

Perspectives on point cloud-based 3D scene modeling and XR presentation within the cloud-edge-client architecture

With the support of edge computing, the synergy and collaboration among central cloud, edge cloud, and terminal devices form an integrated computing ecosystem known as the cloud-edge-client architecture. This integration unlocks the value of data and computational power, presenting significant opportunities for large-scale 3D scene modeling and XR presentation. In this paper, we explore the perspectives and highlight new challenges in 3D scene modeling and XR presentation based on point cloud within the cloud-edge-client integrated architecture. We also propose a novel cloud-edge-client integrated technology framework and a demonstration of municipal governance application to address these challenges.

#4345 NOVEL REAL-TIME PREDICTION SYSTEM FOR HYPOTENSION IN DIALYSIS COMBINING RISK INCIDENCE AND NUMERICAL PREDICTION MODEL

Abstract Background and Aims Although there have been many studies on the machine learning prediction model of intradialytic hypotension, the high prediction accuracy in the academic field needs to be transferred to the clinical environment to aid in medical professionals' decision-making process. Before this clinical trial, our team built a data storage, edge computing, and decision support system. However, to achieve real-time decision support, real-time data transmission is also needed along with the system to achieve clinical decision needs. Therefore, this study aims to integrate the AI predictive model with edge computing to build an overall decision support system for intradialytic hypotension in the real-world clinical environment; we have developed a system architecture to deploy the system. Method Development of the IDH decision support system: We developed a complete IDH prediction system to be implemented in dialysis centers and hospitals, which incorporates patient data retrieval to AI prediction models to the web application and monitoring dashboard for medical professionals. The system architecture includes a bridge service used to retrieve the HIS database (patient history), DIS database (dialysis information), and NIS database (nursing care). The data is stored in a data warehouse and then converted into a structured format for AI model training and prediction. Furthermore, an API Server, a communication service for the database and front-end web, is used to present the warning signals in real-time. Lastly, there is a multi-bed monitoring dashboard presenting personalized patient information, the incidence rate of hypotension and numerical prediction of the actual blood pressure value. Design and validation of the online AI prediction model The real-time AI prediction model is built through automatic machine-learning training, which retrains the model every week, utilizing the data of the past 180 days. The features of the input data include biochemistry lab data, nursing records, demographical characteristics, hemodialysis machine parameters and vital signs; it then uses the SMOTE method to fill in missing values and via RFECV to do feature extraction and selection. Lastly, the model is trained through improved RNN models of LSTM (Long short-term memory) and GRU (Gated Recurrent Unit) integrated with machine learning models of XGboost and LightGBM algorithms to generate a combination of binary classification and numerical prediction. Hypotension's output label is defined when systolic blood pressure drops below 90 mmHg. Finally, the validation of the AI machine learning models for binary classification used accuracy score, specificity and sensitivity, whereas the numerical prediction model used RMSE and MAE. Results With the implementation of the edge computing and decision support system, clinical and dialysis data of the patients upload directly to the data warehouse in real-time, and it presents the hypotension prediction information to the medical professional's computer. As a result, the online real-time AI prediction system for the binary classification model yielded an accuracy score, specificity and sensitivity of 0.95, 0.99 and 0.86, respectively, and the RMSE and MAE yielded 17.5 and 11.6, respectively. Furthermore, the model predicted within 15 seconds to ensure a real-time and automatic parameters update. Conclusion Introducing and implementing this integrated intelligence AI prediction system reduces the occurrence of hypotension in dialysis, improves patient safety and satisfaction of patients, and enhances the work environment of medical professionals. Furthermore, this framework and validation can also be provided to home dialysis patients to improve their quality of care. In addition, this system architecture can also be extended to other medical conditions in developing intelligent decision support systems for practical clinical applications.

Research on university laboratory network security based on Cloud Computing

Abstract As information technology increasingly develops, cloud computing has become an emerging computer network technology, which brings great convenience to network life. In order to ensure network security while fully enjoying this convenience, this paper established the laboratory cloud platform and network topology using the single node deployment mode of Openstack, obtained the software system architecture by applying B/S three-tier architecture model for laboratory development, and finally constructed cloud computing system of the laboratory by integrating encryption technology, digital signature technology, fingerprint recognition technology and dynamic password technology. After functionality, security and performance tests of the laboratory cloud platform, the application effect indicates that, some functional modules of the platform are in line with the desired results, the operation success rate is up to 100%, a 64G server can assume the operation of about 22 virtual machines, the number of server receiving and processing reaches 33 per second, and the throughput climbs up to 980,000 bytes per second. Therefore, it is concluded that with the support of cloud computing, the laboratory basically meets the desired use demand, which can effectively intercept unauthorized operation instructions, and get good carrying capacity and load capacity while strengthening network security.

Soft querying powered by user-defined functions in J-CO-QL+

Soft querying on databases (i.e., selecting data items that partially match selection conditions) was investigated on top of classical relational databases in past research works; however, constraints and limitations posed by relational DBMSs significantly limited the practical effects of this research. The advent of JSON as the format for representing and sharing data over the Internet, together with the birth of JSON document stores (a specific category of NoSQL databases), is now changing the panorama. In fact, the need to integrate and query large JSON data sets is now calling for novel and powerful tools for managing and integrating JSON data sets in a flexible way. At the University of Bergamo (Italy), we are devising the J-CO Framework, which is a platform-independent tool that relies on a high-level and general-purpose language named J-CO-QL+: among all its features, it provides capabilities towards “soft querying” of JSON documents.However, a general-purpose language, although extremely powerful, cannot provide support for domain-specific computations that often relies on procedural algorithms. In this paper, we show how supporting user-defined functions actually empowers J-CO-QL+ users towards applying soft querying on JSON data sets. User-defined functions written both in JavaScript and in Java are accepted by the J-CO-QL+Engine: in this paper, we present how to define them and the different execution performance.

INFLUÊNCIAS DE APLICATIVOS DE EXERCÍCIO FÍSICO SOBRE A SAÚDE DE ESTUDANTES UNIVERSITÁRIOS OBESOS

ABSTRACT Introduction In a scenario of continuous improvement of telecommunications network infrastructure and the maturation of related industrial chains, sports-oriented cell phone applications can meet the needs of different types of user groups due to their characteristics - close to the human body, able to collect users’ health, favorite sports, location, and other personal characteristic data in real-time. With the support of cloud computing, big data, and other techniques, the regularity of users’ sports health can be analyzed. Objective Explore the effect of obese college students’ use of cell phone apps on their exercise routines. Methods 53 obese college students volunteered for the research project. They were investigated by a questionnaire about their use of apps to aid exercise and health. Then, the participants’ exercise status was monitored, including weekly exercise time, exercise frequency, and exercise items, among other pertinent data. Finally, physical health data collected before and after the experiment were analyzed and statistically compared. Results Some obese college students did not acquire the habit of adhering to physical exercise, their psychological resilience and ability to fight frustration proved deficient, and the effect of an exercise assisted by cell phone apps left to be desired. However, for some obese college students who determined to exercise, the regular moderate exercise showed an improvement in physical health. Conclusion Exercise-focused health maintenance apps can stimulate obese college students’ enthusiasm for exercise, making the exercise process more enjoyable, controlled, and scientific. Level of evidence II; Therapeutic studies - investigation of treatment outcomes.