QML ARENA v1.0
RATIONALE
Developers:
Dr. Paolo Bonfini, Napoleon Papoutsakis, Anastasios Litsas, Nikolaos Schetakis, Dr. Alessio Di Iorio

Introduction
Machine Learning (ML) has rapidly become one of the most transformative technologies of our time, reshaping industries and enabling new possibilities across a broad spectrum of domains. From predictive analytics and natural language processing to computer vision and personalized recommendations, ML applications are integral to everything from scientific research and system automation to healthcare diagnostics and financial modeling. In a world increasingly driven by data, the ability to exploit ML tools effectively has become the key competitive advantage, pushing boundaries and driving innovation.
However, as demand for ML applications continues to grow, there remains a critical challenge: educating and training a skilled workforce capable of effectively conceiving, building, deploying, and maintaining these solutions. For both students aspiring to break into the field, and trainers working to equip them with the right skills, a comprehensive understanding of ML tools, best practices, and the rapidly evolving technology landscape is essential. Yet, access to focused, coherent, and practical learning experiences remains a significant bottleneck, hindered by fragmented instruction and the lack of comprehensive, guided educational tools.
This challenge becomes even more pronounced in emerging field of Quantum Machine Learning (QML), an area poised to offer unprecedented computational advantages through the power of quantum computing. Despite its potential, QML remains an unfamiliar and somewhat abstract concept to most practitioners. The complexity of quantum algorithms and the nascent state of quantum hardware present significant educational hurdles. As such, a platform dedicated to guiding users through both classical and quantum ML, while simplifying the steep learning curves involved, fills a vital need.
Our platform addresses these challenges head-on by serving as a bridge between current ML practices and the emerging frontier of QML. By offering structured guidance and resources, it aims to empower trainers, students, and professionals to navigate and master these rapidly evolving domains, positioning them at the forefront of technological transformation.

Key Features
While many practitioners may possess deep technical knowledge of individual models or algorithms, they often lack the practical framework needed to design and execute a complete, end-to-end ML workflow.

Our platform is designed to provide a comprehensive, user-centric experience in building and refining ML and QML models. By addressing each crucial stage of a ML pipeline, it offers a suite of tools that simplify and enhance the user’s journey from data exploration to model evaluation. Below is an overview of the key features offered by the platform.

Data Visualization
Effective ML projects begin with a thorough understanding of the data. Our platform provides robust data visualization tools that empower users to explore, interpret, and gain meaningful insights from their datasets. Through interactive plots, charts, and graphs, users can quickly identify trends, outliers, and distributions, facilitating a more intuitive grasp of data characteristics.

Data Preprocessing
The importance of data preprocessing cannot be overstated, as it often determines the success of a model. Our platform guides users through essential preprocessing steps, including handling missing values, normalizing or scaling features, encoding categorical variables, and more. These automated yet customizable tools allow for fine-tuned adjustments that enhance data quality, reduce noise, and align features for optimal modeling outcomes.

Modeling
Users can seamlessly move from data preparation to modeling, building powerful ML and QML models through an intuitive interface. With built-in support for popular ML libraries and frameworks, users can experiment with diverse flavors of deep learning layers (e.g., fully convolved, recurrent layers, etc.). Similarly, it is possible to readily set up quantum layers by selecting their basic properties (e.g., number of qubits, embedding, entanglement, etc.). The platform simplifies model construction and configuration, allowing users to bring their ideas to life with minimal barriers. The model assembly takes place “behind the scenes,” ensuring that the user does not need to deal with issues such as inconsistencies between connected layers, inappropriate activation function choices, and similar issues that are particularly challenging for the novice.

Pipeline Creation
Our platform offers a flexible pipeline creation tool, enabling users to design, implement, and review end-to-end ML workflows with ease. From data ingestion and preprocessing to model training, validation, and deployment, each stage of the pipeline is streamlined, allowing users to build complex solutions without getting lost in tedious manual processes. This feature ensures reproducibility, efficiency, and rapid iteration through different possible ML solutions.

Exploration of Results
Once a model is trained, the platform’s exploration tools allow users to dive deep into model results, providing detailed metrics, visualizations, and insights. This feature supports error analysis and performance tracking, helping users identify strengths and weaknesses within their models and make data-driven decisions for further improvements.
Comparison of QML Models with Classical Ones
As quantum machine learning continues to evolve, comparing its performance against classical approaches is vital to understanding its potential and limitations. Our platform offers a dedicated module for comparing QML models with their classical counterparts, as well as a set of traditional ML algorithms (e.g., SVM, Random Forest), providing detailed analysis on computational efficiency, accuracy, and overall performance metrics. This feature enables users to make informed decisions about the applicability of QML in various use cases and track the emerging advantages offered by quantum computing.

Educational Aspects
Our platform is designed with a strong emphasis on educational innovation and comprehensive coverage of machine learning concepts, making it an essential tool for both students and trainers. With a focus on guiding users from foundational principles to advanced topics, it supports a wide range of tasks and methodologies, ensuring a well-rounded and interactive learning experience.

Support for a Variety of Tasks
The platform covers a broad spectrum of tasks, including regression and classification, allowing users to explore different types of modeling challenges. Whether predicting numerical outcomes, identifying categories, or exploring more complex relationships, the platform ensures that users have access to the tools they need to build models that align with their goals.

Fully Interactive Components
Every aspect of the platform is designed with interactivity in mind. From data visualization and preprocessing steps to pipeline creation and model evaluation, users can engage with each component dynamically. This hands-on approach accelerates learning and deepens understanding, as users can make changes, observe outcomes in real time, and gain insights through direct experimentation. The interactive environment encourages exploration, promotes active learning, and enhances retention far more effectively than traditional teaching methods based solely on hard coding tutorials.

Rich Visualizations
Understanding complex data and model behavior becomes easier through extensive visualization capabilities. Our platform offers a wealth of charts, graphs, and other visual tools to illustrate data distributions, model performance, and error analysis. This wealth of visual feedback allows users to intuitively grasp important concepts and spot trends, outliers, and relationships within their data, providing an unparalleled depth of understanding.

Teaching of Best Practices
The platform places special emphasis on teaching best practices, which are often overlooked or misunderstood by many practitioners. It guides users through essential methodologies such as train/test data splitting and cross-validation, ensuring they understand how to properly evaluate model performance and prevent issues like overfitting. By embedding these practices directly into the workflow, the platform helps users develop habits that lead to robust, reliable models and reproducible results.

Automatic Consistency Checks
Building effective machine learning models requires consistency and data integrity. To address this need, the platform includes automatic consistency checks that help users detect potential issues, such as data leakage or incorrectly formatted inputs. By catching these common pitfalls early in the development process, users are better equipped to build strong, reliable solutions.

AI-Powered Suggestion Engine
One of the standout features of the platform is its AI-powered suggestion engine based on language models, which provides contextual, real-time feedback and corrections to users as they build models. This intelligent assistant helps identify errors, recommends fixes, and offers guidance on optimizing model configurations. By acting as an always-available expert, it accelerates learning and ensures that users gain a deeper understanding of model creation, validation, and deployment without becoming frustrated by common mistakes.
With its focus on comprehensive coverage, interactivity, visual engagement, and real-time guidance, our platform delivers a truly innovative and all-encompassing learning experience. It empowers users to gain proficiency in machine learning, instilling not only knowledge but also the practical skills necessary for mastery in classical and quantum machine learning domains.

Exportable, Interpretable Code
One of our key future goals is to enable users to export interpretable, high-quality code (in the form of jupyter notebooks) that allows for independent work beyond the platform. This feature empowers learners and professionals alike to deepen their understanding, customize solutions, and apply their knowledge outside the platform’s interface. By generating clean, well-documented code tailored to each unique project, we ensure that users are not only learning concepts but also building transferable, practical skills.

Innovation and Purpose-Driven Learning
While numerous platforms—both open-source and commercial—already exist for automated machine learning (AutoML), such as Auto-sklearn, H2O, and Google Cloud AutoML, they often prioritize automation at the expense of transparency. These solutions, while efficient, typically operate as “black boxes,” obscuring the underlying processes and preventing users from gaining a deep understanding of model development. Moreover, they lack a true educational focus, often leaving users with little grasp of critical concepts and best practices.
While there are several web-based solutions, such as TensorFlow Playground, designed for educational purposes, these platforms often focus solely on isolated aspects of machine learning, such as building and visualizing simple models. Although they provide a valuable introduction to ML concepts, they lack the comprehensive scope needed to guide users through the entire process of data analysis, modeling, and evaluation.
In contrast, our platform breaks from this norm by prioritizing both user engagement and learning, offering full interactivity, detailed explanations, and a guided experience that empowers users to comprehend each step of the ML and quantum ML workflow. This emphasis on understanding, rather than simply automating, makes our platform a uniquely powerful educational tool. To our knowledge, it is the only platform that encompasses the end-to-end process.

Support for Time Series Tasks
Time series analysis represents a critical area for many practical applications in finance, healthcare, energy, and more. We aim to add robust support for time series tasks, including forecasting, trend detection, and anomaly analysis, providing users with the tools needed to tackle this complex yet highly valuable domain.

Probabilistic Models
The addition of probabilistic modeling will offer a powerful way for users to incorporate uncertainty into their analyses, making predictions more realistic and actionable. Support for probabilistic approaches will enable users to make decisions that factor in the inherent uncertainty of data, offering deeper and more flexible insights.

Cloud Deployment
Cloud deployment is one of the more ambitious goals in our roadmap, requiring substantial effort and investment. This enhancement would allow users to access our platform from anywhere, manage data storage efficiently, and execute multiple models concurrently in a distributed environment. This feature will provide scalable performance, accommodating users with both small and large datasets, and ensure reliable, secure handling of data and resources.

Account-Based Usage and Multi-User Support
We plan to introduce account-based usage, facilitating a personalized experience for multiple users. With individual user profiles, teams and individuals will be able to access their projects, track their progress, collaborate, and securely manage data and models. This capability will promote a more tailored and cohesive user experience, encouraging greater engagement and collaboration.

Broadening Our Suite of Modern Models
By incorporating a larger set of modern, cutting-edge models, users will have access to a diverse and up-to-date toolkit. This expanded selection will empower users to explore a wider range of approaches, from classical techniques to state-of-the-art advancements, allowing for greater creativity and versatility in their projects.

Refinement of the AI Suggestion Engine
Our AI-powered suggestion engine has already proven to be a powerful tool in guiding users, providing real-time feedback and correcting errors during model creation. We aim to further refine and enhance this engine, making it even more intelligent, context-aware, and capable of delivering targeted advice. Improvements will include more precise diagnostics, adaptive recommendations based on user behavior, and greater customization options to meet individual learning needs. This refinement will empower users to learn and correct mistakes faster, deepening their expertise and confidence.

By introducing these new features, we aim to transform our platform into a more comprehensive, powerful, and user-centric solution for machine learning and quantum machine learning education. With the support of additional funding, we can ensure our platform meets the diverse needs of users, remains on the cutting edge of technology, and continues to make a meaningful impact in the world of QML education.

Demonstration of Our Quantum Machine Platform

Figure 1. Demonstration of Our Quantum Machine Learning Platform The tabs “Data & Task” (A), “Preprocessing” (B), “Modeling” (C), “Pipeline” (D), “Run Fit” (E), and “Results” (F) guide users through the complete analysis workflow.
A. Users can select their own datasets or utilize one of the built-in “toy” datasets to explore the platform’s capabilities. In this demonstration, we have chosen the “moons” dataset. Users can also specify the type of task — here, “classification” was selected.
B. A variety of preprocessing tools and their configurable parameters are available. These steps can be sequentially applied, allowing for flexible transformations of the data. In this example, a “Standard Scaler” was selected, which could be followed by additional processes like Principal Component Analysis (PCA) for dimensionality reduction. A warning message ensures users are reminded that preprocessing steps are applied solely to the training set to prevent data leakage.
C. The platform enables the user to build a sequential Neural Network (NN) model layer by layer, with customizable parameters and global settings such as the learning rate. Input and output layers are automatically set based on the selected data and task from step A. Users can define classical and/or quantum layers at this stage thus forming a Hybrid (Classical-Quantum) Neural Network. Users can track the network architecture through a graphical visualization and an accompanying table. In this case, we constructed a NN with three hidden layers, with the final layer being a quantum layer containing four qubits. When quantum layers are used, the platform generates a “Classical Equivalent” model, substituting the quantum components with their classical counterparts (here, a “Dense” layer with 2^n_qubits neurons). This serves as a benchmark for comparison. The system also automatically verifies compatibility between layers, issuing warnings if any inconsistencies are detected, and users can access tips via a “Get Tip” button powered by a fine-tuned LLM for tailored advice.
D. Users can select a training and evaluation protocol. For this demonstration, we chose a simple “train/validation/test” split, though Cross-Validation could also be selected.
E. By pressing the “Fit your model” button, users can observe the loss curves over different epochs for both the User’s model and the Classical Equivalent. Since this is a classification task, accuracy metrics are also displayed to illustrate performance changes throughout training.
F. After model fitting, several diagnostic tools and performance metrics are presented to benchmark the classical and the Hybrid NN. This includes standard performance measures evaluated on the test set (top), a Receiver Operating Characteristic (ROC) curve (middle), and visualizations of the decision functions produced by both the User’s and Classical Equivalent models (bottom). The top bar also offers users the option to adjust the probability threshold used to separate classes (defaulting to 50%), allowing for dynamic exploration of how various metrics respond to changes in this threshold.

Figure 1. Demonstration of Our Quantum Machine Learning Platform The tabs “Data & Task” (A), “Preprocessing” (B), “Modeling” (C), “Pipeline” (D), “Run Fit” (E), and “Results” (F) guide users through a complete analysis workflow.
Additional Resources
Online App: QML Arena (accessed July 2025)
Explore and interact with quantum machine learning models directly in your browser.
Video Tutorial: QML Arena – Getting Started Guide (accessed July 2025)
Step-by-step walkthrough of the platform’s features and workflows.

Cite as
Bonfini, P., Papoutsakis, N., Litsas, A., Schetakis, N., & Di Iorio, A. (2025). Quantum Machine Learning (QML) Arena (1.0). Zenodo. https://doi.org/10.5281/zenodo.15796747

License
QML Arena is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).
You are free to use, share, and adapt QML Arena for non-commercial educational and academic research purposes, provided that you:
Credit Quantum Innovation IKE and the original developers
Include a proper citation in any work that uses or builds upon the platform
Commercial use, redistribution, or integration into proprietary software is not permitted without prior written consent. Full license text: https://creativecommons.org/licenses/by-nc/4.0/
References
1.Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., & Lloyd, S. (2017). “Quantum Machine Learning.” Nature, 549(7671), 195–202.
2.Mari, A., Bromley, T. R., Izaac, J., Schuld, M., & Killoran, N. (2020). “Transfer Learning in Hybrid Classical-Quantum Neural Networks.” Quantum, 4, 340.
3.Benedetti, M., Lloyd, E., Sack, S., & Fiorentini, M. (2019). “Parameterized Quantum Circuits as Machine Learning Models.” Quantum Science and Technology, 4(4), 043001.
4.Schetakis, N., Aghamalyan, D., Griffin, P. et al. Review of some existing QML frameworks and novel hybrid classical–quantum neural networks realising binary classification for the noisy datasets. Sci Rep 12, 11927 (2022). https://doi.org/10.1038/s41598-022-14876-6
5.Schetakis, N.; Aghamalyan, D.; Boguslavsky, M.; Rees, A.; Rakotomalala, M.; Griffin, P.R. Quantum Machine Learning for Credit Scoring. Mathematics 2024, 12, 1391. https://doi.org/10.3390/math12091391
6.Blazakis, K.; Katsigiannis, Y.; Schetakis, N.; Stavrakakis, G. One Day Ahead Wind Speed Forecasting based on Advanced Deep and Hybrid Quantum Machine Learning. In Proceedings of the Springer 1st International Conference on Frontiers of Artificial Intelligence, Ethics and Multidisciplinary Applications (FAIEMA 2023), Athens, Greece, 25–26 September 2023. https://doi.org/10.1007/978-981-99-9836-4_13
7.Quantum neural networks with data re-uploading for urban traffic time series forecasting, Schetakis, N., Bonfini, P., Alisoltani, N. et al. Scientific Reports, 15, 19400 (2025). https://doi.org/10.1038/s41598-025-04546-8
8.Power Theft Detection in Smart Grids Using Quantum Machine Learning
K. Blazakis, N. Schetakis, M. M. Badr, D. Aghamalyan, K. Stavrakakis, G. Stavrakakis. IEEE Access, vol. 13, pp. 61511–61525, 2025. https://doi.org/10.1109/ACCESS.2025.3558143

Appendix
Hybrid Neural Networks (HNNs), which integrate classical neural network layers with quantum neural layers, have been explored as a promising approach to leverage the unique strengths of both classical and quantum computing. In an HNN, a quantum neural layer is sandwiched between classical layers, allowing the model to capitalize on quantum computing’s potential for high-dimensional data processing and complex entanglement, while retaining the familiarity, robustness, and scalability of classical neural network structures. This hybrid structure is not only theoretically appealing but has also shown empirical advantages in various studies [1,2,3] while members of the consortium (QUINN, QUIBTECH) have validated this technology for energy and finance problems [4,5,6,7,8]. The quantum layer can use either a classical pc as backend or a quantum device (i.e. IBM Quantum) .

Figure 2. Block diagram of the Hybrid Neural Network, featuring a quantum layer (Variational Classifier with integrated Data Re-upload Technique [4]; VC-DRC) placed between classical neural layers.

Key Educational Features:
●Interactive Components: Engage with data, models, and results for active, hands-on learning.
●Rich Visualizations: Charts and graphs illustrate data distributions, model behaviors, and performance metrics.
●Best Practices in Teaching: Emphasis on methodologies like data splitting and cross-validation ensures robust models.
●Automatic Consistency Checks: Early detection of issues like data leakage ensures reliable model development.
●AI-Powered Suggestion Engine: Real-time feedback helps users optimize configurations, learn quickly, and build confidence.
●Benchmarking Classical vs. Quantum Neural Networks
QML platform enables users to build their own Hybrid Neural Networks (HNNs) and benchmark them against classical models with similar architectures. This feature allows learners to:
A.Build Hybrid Models: Users can design and customize their own hybrid models by integrating quantum layers within classical neural network structures.
B.Compare the performance of hybrid models against classical counterparts using identical architectures to evaluate the impact of HNNs.
C.Explore and analyze the output of both the classical and quantum models, gaining insights into how each approach handles a given task.
D.Running Time & Efficiency: Evaluate the running time and computational efficiency of quantum-enhanced models versus purely classical models.
E.Access a wide range of performance metrics, such as accuracy, loss, training time, and resource usage, to assess the strengths and limitations of each model.
Unlike many solutions that prioritize automation as “black boxes,” our platform emphasizes transparency and education. While tools like TensorFlow Playground introduce ML concepts, our solution offers greater depth, interactivity, and guidance for mastering both classical and quantum ML.
Planned Future Enhancements:
●Exportable, Interpretable Code: Generate customizable code for independent exploration.
●Support for Time Series Tasks: Tools for forecasting and trend detection.
●Probabilistic Models: Improved insights through predictions with incorporated uncertainty.
●Cloud Deployment: Scalable access and multiple user support.
●Expanded Model Suite: A broader selection of modern models and techniques.
●Refined AI Suggestion Engine: Improved precision and tailored recommendations.
Our comprehensive, user-centric platform bridges classical and quantum ML education. By emphasizing interactive, practical learning, we equip users with foundational knowledge and hands-on experience to drive innovation in a rapidly evolving landscape.