Abstracts-QAI-EN

a:3:{s:6:"locale";s:5:"en_US";s:3:"rtl";i:0;s:9:"flag_code";s:2:"us";}
Determining links in product data using Quantum Restricted Boltzmann Machines

Determining links in product data using Quantum Restricted Boltzmann Machines

Abstract:

The increasing share of software in products not only drives innovation but also increases complexity. To minimize the risk of malfunctions in software-intensive products and ensure traceability, links between development stages and production data are necessary. These are mandated by regulations such as ISO/IEC 15288 and DIN/ISO 26262. The Digital Data Package standard enables the management of such links. However, implicit links currently have to be created manually, which leads to challenges due to the scope and frequent product changes. A promising approach for the automatic identification of links is the use of classifiers. In particular, Quantum Restricted Boltzmann Machines offer a viable solution due to the limited availability of linked development data and their high susceptibility to noise. For evaluation, classical neural networks and pre-trained classifiers are used. As established methods in pattern recognition, they serve as a baseline for assessing new classifiers.

Author:

Simon Hehnen

Advisors:

Michael Kölle, Jonas Stein, Dr. Fabrice Mogo Nem (PROSTEP AG), Claudia Linnhoff-Popien

Student Thesis | Published April 2025 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Emerging Cooperation through Quantum Entanglement in Multi-Agent Systems

Emerging Cooperation through Quantum Entanglement in Multi-Agent Systems

Abstract:

This thesis investigates the feasibility of quantum entanglement enhancing cooperation in multi-agent reinforcement learning. Utilizing the iterated prisoner dilemma as a benchmark, we propose a decentralized multi-agent reinforcement learning framework where two MARL agents equipped with variational quantum circuits can affect each other through quantum entanglement. Unlike existing approaches that rely on dedicated quantum communication channels, this thesis examines whether entanglement alone can facilitate cooperative equilibria. Therefore, we evaluate the effects of different entanglement architectures to develop mutual cooperative strategies that escape the Nash equilibrium. The results of our experiments indicate that while entanglement can facilitate strategies that outperform the defection baseline, long-term cooperative behavior remains unfeasible, suggesting that quantum correlations alone are insufficient to sustain cooperative strategies in multi-agent reinforcement learning settings.

Author:

Marvin Heinrich

Advisors:

Michael Kölle, Leo Sünkel, Claudia Linnhoff-Popien

Student Thesis | Published March 2025 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Distributed Quantum Machine Learning -Training and Evaluating a Machine Learning Model on a Distributed Quantum Computing Simulator

Distributed Quantum Machine Learning -Training and Evaluating a Machine Learning Model on a Distributed Quantum Computing Simulator

Abstract:

The training and execution of machine learning models on quantum hardware is typically limited by the number of available qubits. A potential approach to overcoming this limitation is Distributed Quantum Machine Learning (DQML), where models are partitioned and executed across multiple quantum computers. While this increases the number of available qubits and potentially enables the training of larger models, it also introduces substantial classical and quantum communication overhead, leading to increased computational costs and extended training times. To investigate this approach and its limitations, this thesis presents a DQML model using a classical server and two quantum clients, implemented with the distributed quantum framework NetQASM. We evaluated the model on datasets with two and four features using a quantum network simulator and it achieved classification performance comparable to that of a centralized quantum baseline. To address the communication overhead, which resulted in training times of 50 to 500 minutes per epoch, optimizations in circuit design, entanglement generation, and distributed gate execution were implemented and evaluated. These adaptations led to a reduction in runtime of up to 60% while maintaining competitive classification accuracy.

Author:

Kian Izadi

Advisors:

Leo Sünkel, Michael Kölle, Thomas Gabor, Claudia Linnhoff-Popien

Student Thesis | Published March 2025 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


QUBO-Generation for (MAX-)3SAT via generative AI-Methods

QUBO-Generation for (MAX-)3SAT via generative AI-Methods

Abstract:

Creating QUBOs for 3-SAT formulas using pattern QUBOs poses several challenges. Generating pattern QUBOs and building the QUBO structure itself is technically demanding due to the brute-force approach. In this study, two machine-learning approaches for QUBO generation given a 3-SAT formula are tested. Various encoding methods were explored for representing formulas and matrices. Formula encodings included vector, Word2Vec, and BERT-based methods, while latent representations were tested on QUBOs. As an initial model, a conditional autoencoder was used, with variations like dual encoders and pretrained encoders based on a RESNET18 architecture also evaluated. Accurate QUBOs could be generated for formulas with a single clause, but for formulas with up to four clauses, energy levels of solution and non-solution states overlapped. Finally, a conditional diffusion model was implemented and trained on 5 and 7 clause random formulas using vector, Word2Vec, and BERT formula embeddings. QUBOs generated with BERT formula embeddings fulfilled the highest average number of clauses per formula, though most formulas remained unsolved. Training with masked diffusion further improved performance, as QUBOs generated with masking fulfilled, on average, one additional clause. However, this approach requires a predefined mask during data generation. The sparse QUBO data structure and challenges in encoding 3-SAT formulas are likely primary factors behind these results.

Author:

Philippe Wehr

Advisors:

Sebastian Zielinski, Michael Kölle, Claudia Linnhoff-Popien

Student Thesis | Published March 2025 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Evaluating Parameter-Based Training Performance of Neural Networks and Variational Quantum Circuits

Evaluating Parameter-Based Training Performance of Neural Networks and Variational Quantum Circuits

Abstract:

In recent years, neural networks (NN) have played a central role in significant advancements in the field of machine learning. The increasing complexity of machine learning tasks leads to NNs with a growing number of trainable parameters, resulting in high computational and energy demands. Variational quantum circuits (VQC) are a promising alternative. They leverage quantum mechanics to model complex relationships and tend to require fewer trainable parameters compared to NNs. In this work, we evaluate and compare the training performance of NNs and VQCs on simple supervised learning and reinforcement learning tasks, considering multiple models with varying numbers of parameters. The experiments with VQCs are conducted using a simulator. To approximate how long training the VQCs would take using currently available real quantum hardware, selected parts of the training process are executed using a real quantum computer. Our results confirm that VQCs can achieve performance comparable to NNs while requiring significantly fewer parameters. Despite longer training times, our findings suggest that VQCs could be advantageous for certain machine learning tasks, particularly as quantum technology continues to rapidly advance, algorithms are optimized, and VQC architectures are improved.

Author:

Alexander Feist

Advisors:

Michael Kölle, Jonas Stein, Claudia Linnhoff-Popien

Student Thesis | Published January 2025 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Reinforcement learning-supported state preparation using parameterized quantum gates

Reinforcement learning-supported state preparation using parameterized quantum gates

Abstract:

This thesis investigates the application of reinforcement learning (RL) in order to optimize the state preparation in parameterized quantum circuits. By using RL algorithms, an agent is trained to find the optimal sequence of quantum gates so as to reconstruct predetermined target states. Particular attention is paid to the challenges of using parametric gates, which require continuous optimization when compared to discrete circuits. Dierent approaches, including one- and two-stage methods as well as hyperparameter optimizations, are evaluated experimentally. The results show that RL-based methods can successfully contribute to the reduction of circuit depth, however this applies mainly to simple circuits. More complex circuits require deeper adaptations of the optimization strategy in order to achieve similar success.

Author:

Isabella Debelic

Advisors:

Michael Kölle, Philipp Altmann, Claudia Linnhoff-Popien

Student Thesis | Published December 2024 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Comparison of different hybrid quantum machine learning approaches for image classification on quantum computers

Comparison of different hybrid quantum machine learning approaches for image classification on quantum computers

Abstract:

Nowadays, Machine learning (ML) and the classification of images are becoming increasingly important. ML is used amongst others in autonomous vehicles to determine obstacles or in medicine for the automatic detection of diseases. However, the demands on neural networks used for image classification are constantly increasing as the features in the images become more and more complex. A promising solution in this area is quantum computing, or more precisely quantum machine learning (QML). Due to the advantages that qubits used in quantum computers bring with them, QML approaches could achieve significantly faster and better results than conventional ML methods. Quantum computing is currently in the so-called ’noisy intermediate-scale quantum’ (NISQ) era which means that quantum computers only have a few qubits, which are prone to errors. Accordingly, quantum machine learning cannot be easily implemented. The solution are hybrid approaches that use classical structures and combine them with quantum circuits.

This work analyzes the hybrid approaches Quanvolutional Neural Network (QCNN), Quantum Transfer Learning (QTL) and Variational Quantum Circuit (VQC). These are trained to classify the images of the MNIST data set. The training is takes place several times with different seeds in order to test the robustness of the approaches. They are then compared based on accuracy, loss and training duration. Additionally, a conventional Convolutional Neural Network (CNN) is used for comparison. Finally, the most efficient approach will be determined. The evaluation of the experiment shows that the QCNN achieves significantly better results than QTL and VQC. However, the conventional CNN performs better than the QCNN in all metrics.

Author:

Nicolas Holeczek

Advisors:

Leo Sünkel, Philipp Altmann, Claudia Linnhoff-Popien

Student Thesis | Published December 2024 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Architectural Influence on Variational Quantum Circuits in Multi-Agent Reinforcement Learning: Evolutionary Strategies for Optimization

Architectural Influence on Variational Quantum Circuits in Multi-Agent Reinforcement Learning: Evolutionary Strategies for Optimization

Abstract:

The field of Multi-Agent Reinforcement Learning (MARL) is becoming increasingly relevant in domains that involve the interaction of multiple agents, such as autonomous driving and robotics. One challenge in MARL is the exponential growth of dimensions in the state and action spaces. Quantum properties o!er a solution by enabling compact data processing and reducing trainable parameters. One drawback of gradient-based optimization methods in Quantum MARL is the possibility of Barren Plateaus impeding effective parameter updating, thereby hindering convergence. Evolutionary Algorithms, however, bypass this issue as they do not rely on gradient information. Building on research that demonstrates the potential of Evolutionary Algorithms in optimizing Variational Quantum Circuits for MARL tasks, we examine how introducing architectural changes into the evolutionary process affects optimization. We explore three different architecture concepts for Variational Quantum Circuits — Layer-Based, Gate-Based, and Prototype-Based — by applying two evolutionary strategies: one involving both recombination and mutation (ReMu), and the other using mutation only (Mu). To evaluate the efficacy of these approaches, we tested them in the Coin Game, comparing them to a baseline without architectural modifications. The mutation-only strategy with the Gate- Based approach yielded the best results, achieving the highest scores, number of coins collected, and own coin rates while using the fewest parameters. Furthermore, a variant of the Gate-Based approach with results comparable to those of the baseline required significantly fewer gates, resulting in an acceleration of the runtime by 90.1%.

Author:

Karola Schneider

Advisors:

Michael Kölle, Leo Sünkel, Claudia Linnhoff-Popien

Student Thesis | Published November 2024 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


The Trainability of Quantum FederatedLearning

The Trainability of Quantum Federated Learning

Abstract:

This thesis explores the implementation and evaluation of Quantum Federated Learning (QFL), where Variational Quantum Circuits (VQCs) are collaboratively trained across multiple quantum clients. The primary focus is on comparing the performance and trainability of QFL with traditional non-federated quantum machine learning approaches using the MNIST dataset. Experiments were conducted with 2, 3, 4, and 5 clients, each processing different subsets of data, and with varying numbers of layers (1, 2, and 4) in the quantum circuits. The trainability of the models was assessed through the evaluation of accuracy, loss, and gradient norms throughout the training process. The results demonstrate that while QFL enables collaborative learning and shows significant improvements in these metrics during training, the baseline models without federated learning generally exhibit superior performance in terms of final accuracy and loss due to the uninterrupted optimization process. Additionally, the impact of increasing the number of layers on training stability and performance was examined.

Author:

Sina Mohammad Rezaei

Advisors:

Leo Sünkel, Thomas Gabor, Tobias Rohe, Claudia Linnhoff-Popien

Student Thesis | Published November 2024 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Investigating the Lottery Ticket Hypothesis for Variational Quantum Circuits

Investigating the Lottery Ticket Hypothesis for Variational Quantum Circuits

Abstract:

Quantum computing is an emerging field in computer science that has made significant progress in recent years, including in the area of machine learning. Through the principles of quantum physics, it offers the possibility of overcoming the limitations of classical algorithms. However, variational quantum circuits (VQCs), a specific type of quantum circuits utilizing varying parameters, face a significant challenge from the barren plateau phenomenon, which can hinder the optimization process in certain cases. The Lottery Ticket Hypothesis (LTH) is a recent concept in classical machine learning that has led to notable improvements in neural networks. In this thesis, we investigate whether it can be applied to VQCs. The LTH claims that within a large neural network, there exists a smaller, more efficient subnetwork (a “winning ticket”) that can achieve comparable performance. Applying this approach to VQCs could help reduce the impact of the barren plateau problem. The results of this thesis show that the weak LTH can be applied to VQCs, with winning tickets discovered that retain as little as 26.0% of the original weights. For the strong LTH, where a pruning mask is learned without any training, we found a winning ticket for a binary VQC, performing at 100% accuracy with 45% remaining weights. This shows that the strong LTH is also applicable to VQCs. These findings provide initial evidence that the LTH may be a valuable tool for improving the efficiency and performance of VQCs in quantum machine learning tasks.

Author:

Leonhard Klingert

Advisors:

Michael Kölle, Julian Schönberger, Claudia Linnhoff-Popien

Student Thesis | Published November 2024 | Copyright © QAR-Lab
Direct Inquiries to this work to the Advisors


Page 1 of 3